University of Southern California Dissertations and Theses

A radio frequency based indoor localization framework for supporting building emergency response operations

(USC Thesis Other)

A radio frequency based indoor localization framework for supporting building emergency response operations

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

A Radio Frequency Based Indoor
Localization Framework for Supporting
Building Emergency Response Operations

Nan Li

Submitted in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
in
Civil Engineering

Committee members:
Professor Burcin Becerik-Gerber (advisor)
Professor Lucio Soibelman
Professor Bhaskar Krishnamachari

University of Southern California
Los Angeles, California
May 2014
ii

Abstract
Building emergencies especially structure fires are big threats to the safety of building
occupants and first responders. When emergencies occur, unfamiliar environments are
difficult and dangerous for first responders to search and rescue, sometimes leading to
secondary casualties. One way to reduce such hazards is to provide first responders with
timely access to accurate location information. Despite its importance, access to the
location information at emergency scenes is far from being automated and efficient. This
thesis assesses the value of location information through a card game, and identifies a set
of requirements for indoor localization through a survey. The most important five
requirements are: accuracy, ease of on-scene deployment, resistance to damages,
computational speed, and device size and weight. The thesis introduces a radio frequency
(RF) based indoor localization framework to satisfy these requirements. When no
existing sensing infrastructure is accessible in a building and an ad-hoc sensor network
needs to be established, an environment aware beacon deployment (EASBL) algorithm is
developed for supporting a sequence based localization schema. The algorithm is
designed to achieve dual objectives of improving room-level localization accuracy and
reducing the effort required to deploy the ad-hoc sensor network. When there is existing
sensing infrastructure in the building, an iterative maximum likelihood estimation (IMLE)
localization algorithm is developed for the framework. The algorithm integrates a
maximum likelihood estimation technique for location computation. The algorithm also
introduces an iterative process that mitigates impacts of radio signal’s multipath and
iii

fading effects on localization accuracy. Moreover, building information models are
integrated to both algorithms. Building information plays an important role in mitigating
multipath and fading effects in iterative location computation, enabling the metaheuristic
based search for building-specific satisfactory beacon deployment plans, and providing a
graphical interface for user interaction and result visualization. The framework was
validated in both simulation and field tests. The simulation involved two fire emergency
scenarios in an office building, and reported room-level accuracies of above 87.0% and
coordinate-level accuracies of above 1.78 m for the EASBL, and room-level accuracies
of above 95.0% and coordinate-level accuracies of above 0.84 m for the IMLE. The field
tests involved the same test bed and scenarios, and used a smartphone based prototype
that implemented the framework. The field tests reported room-level accuracies of above
82.8% and coordinate-level accuracies of above 2.29 m for the EASBL, and room-level
accuracies of above 84.6% and coordinate-level accuracies of above 2.07 m for the IMLE.
The framework also reduced the deployment effort of ad-hoc sensor networks by 32.1%,
was proven to be robust against partial loss of devices, and could promisingly satisfy
other aforementioned requirements for indoor localization at building emergency scenes.
iv

Acknowledgments
This thesis would not have been possible without the help of many wonderful people with
whom I have had the pleasure to work throughout the past five years. First and foremost,
deepest gratitude is due to my advisor, Professor Burcin Becerik-Gerber, and chair and
co-advisor, Professor Lucio Soibelman. They have always been my invaluable sources of
inspiration and encouragement. Without their moral and academic support, I would not
have been able to overcome the challenges I faced during this journey.
I would also like to thank the other members of my defense and qualifying exam
committees, Professor Bhaskar Krishnamachari, Professor Erik A Johnson and Professor
Milind Tambe, for the enlightening guidance they have provided.
Special thanks to first responders from the Los Angeles and Vernon City Fire
Departments, for dedicating so much of their valuable time and providing many
insightful opinions and advice for this research. I especially thank Assistant Chief
Andrew S. Guth, Battalion Chief Michael Rhodes, Carlos M. Calvillo, and Captain Mark
Curry for the time and assistance they have kindly provided.
I greatly acknowledge the financial support from the University of Southern California
and the National Science Foundation. Thank you to my colleagues, staff and the entire
CEE community at the University of Southern California. You all made my years here in
Los Angeles the best possible.
v

Finally, I would like to express my love and gratitude to my beloved parents, for always
supporting me in every decision I made in life, even if that meant being physically distant.

vi

Table of Contents
Abstract ............................................................................................................................... ii
Acknowledgments.............................................................................................................. iv
Table of Contents ............................................................................................................... vi
List of Figures .................................................................................................................... ix
List of Tables ...................................................................................................................... x
Chapter 1: Introduction ....................................................................................................... 1
1.1 Problem Statement .................................................................................................... 1
1.2 Vision ........................................................................................................................ 3
Chapter 2: Research Objectives and Questions .................................................................. 5
Chapter 3: Review of Building Emergency Response Procedures ..................................... 9
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations..................... 14
4.1 Value of Location Information in Building Emergency Response Operations ...... 14
4.1.1 Card Game ........................................................................................................ 14
4.1.2 Development of Information Items List ........................................................... 17
4.1.3 Card Game Procedures ..................................................................................... 20
4.1.4 Card Game Implementation .............................................................................. 22
4.1.5 Analysis of Card Game Results ........................................................................ 23
4.2 Indoor Localization Requirements for Building Emergency Response Operations 30
4.2.1 Survey Design................................................................................................... 31
4.2.2 Survey Analysis ................................................................................................ 33
4.3 Conclusions ............................................................................................................. 38
Chapter 5: Review and Evaluation of Indoor Localization Technologies ........................ 39
5.1 Review of Indoor Localization for Building Emergency Response Operations ..... 39
5.2 Review of General Indoor Localization Solutions .................................................. 42
5.2.1 Review of INS Based Solutions ....................................................................... 43
5.2.2 Review of AGPS .............................................................................................. 44
5.2.3 Review of Infrared Based Solutions ................................................................. 46
5.2.4 Review of RF Based Solutions ......................................................................... 47
vii

5.3 Evaluation of Indoor Localization Technologies for Building Emergency Response
Operations ..................................................................................................................... 53
5.4 Conclusions ............................................................................................................. 56
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations ......................................................................................................................... 58
6.1 Overview of RF Based Indoor Localization Algorithms ........................................ 58
6.2 The EASBL Algorithm ........................................................................................... 62
6.2.1 Review of the Sequence Based Localization Schema ...................................... 62
6.2.2 Algorithm Design ............................................................................................. 65
6.3 The IMLE Algorithm .............................................................................................. 72
6.3.1 Assumptions ..................................................................................................... 73
6.3.2 Algorithm Design ............................................................................................. 73
6.4 Conclusions ............................................................................................................. 79
Chapter 7: Simulation Based Evaluation of the Localization Framework ....................... 81
7.1 Simulation Setup and Scenarios .............................................................................. 81
7.2 Evaluation of the Framework with No Existing Sensing Infrastructure ................. 85
7.2.1 Evaluation of Metaheuristics ............................................................................ 86
7.2.2 Evaluation of Localization Accuracy and Deployment Effort ......................... 89
7.2.3 Evaluation of Robustness ................................................................................. 93
7.2.4 Tradeoff between Onsite Deployment Effort and Localization Accuracy ....... 94
7.3 Evaluation of the Framework with Existing Sensing Infrastructure ....................... 96
7.3.1 Selection of Fitness Function ........................................................................... 97
7.3.2 Evaluation of Localization Accuracy ............................................................. 100
7.3.3 Evaluation of Robustness against Loss of Transmitters ................................. 103
7.3.4 Evaluation of Robustness against Loss of Transceivers ................................. 104
7.4 Conclusions ........................................................................................................... 106
Chapter 8: Field Test Based Evaluation of the Localization Framework ....................... 107
8.1 Prototype Development ......................................................................................... 107
8.2 Field Test Scenarios, Procedures and Test Bed Setup .......................................... 111
8.3 Evaluation of the Framework with No Existing Sensing Infrastructure ............... 114
8.3.1 Localization Accuracy and Deployment Effort .............................................. 114
8.3.2 Robustness against Partial Loss of Deployed Nodes...................................... 116
viii

8.3.3 Ease of On-Scene Deployment, Computational Speed, and the Size and Weight
of Devices ................................................................................................................ 117
8.4 Evaluation of the Framework with Existing Sensing Infrastructure ..................... 119
8.4.1 Localization Accuracy .................................................................................... 119
8.4.2 Robustness against Partial Loss of Existing Transmitters .............................. 120
8.4.3 Robustness against Partial Loss of Existing Transceivers.............................. 122
8.4.4 Ease of On-Scene Deployment, Computational Speed, and the Size and Weight
of Devices ................................................................................................................ 122
8.5 Conclusions ........................................................................................................... 123
Chapter 9: Discussions .................................................................................................... 124
9.1 Comparison of the Simulation Results and Field Test Results ............................. 124
9.1.1 Performance of the EASBL Algorithm .......................................................... 124
9.1.2 Performance of the IMLE Algorithm ............................................................. 126
9.2 Comparison between the EASBL and the IMLE Algorithms ............................... 129
Chapter 10: Limitations .................................................................................................. 134
Chapter 11: Conclusions ................................................................................................. 138
References ....................................................................................................................... 141

ix

List of Figures
Figure 1: Job titles of the survey respondents ................................................................... 34
Figure 2: Illustration of SBL ............................................................................................. 63
Figure 3: Pseudo code for the iterative process ................................................................ 78
Figure 4: Simulation scenarios.......................................................................................... 84
Figure 5: Illustration of device deployments in the field tests .......................................... 85
Figure 6: Flowchart of EASBL based indoor localization process................................... 86
Figure 7: Convergence speed of three metaheuristics from a sample run in two scenarios
........................................................................................................................................... 89
Figure 8: Accuracy of the EASBL under different variances of signal shadowing .......... 92
Figure 9: Assessment of robustness against partial loss of deployed sensor nodes .......... 94
Figure 10: Tradeoff in scenario 1 and scenario 2 ............................................................. 96
Figure 11: Flow chart of IMLE based indoor localization process .................................. 97
Figure 12: Accuracy of the IMLE under different variances of signal shadowing ........ 103
Figure 13: Assessment of robustness against partial loss of existing transmitters ......... 104
Figure 14: Assessment of robustness against partial loss of existing transceivers ......... 105
Figure 15: Transmitter (a) and smartphone (b) used in the prototype ............................ 108
Figure 16: Interface of the localization application ........................................................ 109
Figure 17: Data flow in the field tests ............................................................................. 111
Figure 18: Assessment of robustness against partial loss of deployed sensor nodes ...... 116
Figure 19: Robustness against partial loss of existing transmitters in the field test ....... 121
Figure 20: Robustness against partial loss of existing transceivers in the field test ....... 122
Figure 21: Comparison of robustness between simulation and field test ....................... 126
Figure 22: Comparison of robustness against loss of transmitters between simulation and
field tests ......................................................................................................................... 128
Figure 23: Comparison of robustness against loss of transceivers between simulation and
field tests ......................................................................................................................... 129
Figure 24: Comparison of the robustness of the two algorithms (simulation) ............... 132
Figure 25: Comparison of the robustness of the two algorithms (field test) .................. 132

x

List of Tables
Table 1: Information Items Used in Building Emergency Response Operations ............. 18
Table 2: Summary of Card Game Results ........................................................................ 27
Table 3: Importance of indoor localization requirements ................................................. 35
Table 4: Best performance of indoor localization technologies against five requirements
........................................................................................................................................... 55
Table 5: Summary of metaheuristics parameter values .................................................... 87
Table 6: Evaluation of localization performance .............................................................. 90
Table 7: Performance of three fitness functions ............................................................... 98
Table 8: Comparison of converged and non-converged estimations ................................ 99
Table 9: Evaluation of the accuracy of the IMLE algorithm .......................................... 101
Table 10: Localization accuracy of the EASBL in the field tests ................................... 115
Table 11: Evaluation of the accuracy of the IMLE algorithm in the field tests.............. 119
Table 12: Localization accuracy of the EASBL ............................................................. 125
Table 13: Localization accuracy of the IMLE ................................................................ 127
Table 14: Comparison of the accuracy between the EASBL and the IMLE .................. 130

Chapter 1: Introduction
1.1 Problem Statement
Building emergencies especially structure fires are big threats to the safety of building
occupants and first responders. For example, public fire departments across the U.S.
attended 480,500 fires in buildings in 2012, which caused 2,380 deaths and 12,875
injuries [1]. When emergencies occur, unfamiliar environments are difficult and
dangerous for first responders to search and rescue, sometimes leading to secondary
casualties. With the increasing number of complex buildings, and less live-fire training,
first responders are twice as likely to die inside structures as they were 20 years ago, and
the leading cause of these line-of-duty deaths is getting lost, being trapped or disoriented
[2]. Statistics show that 87% of fire-related firefighter fatalities and injuries occur in
structure fires [3]. A total of 159 firefighters died between 2000-2011 in the U.S. when
responding to structure fires, one major cause of which was firefighters getting lost [4,5].
One way to reduce such hazards is to provide firefighters with timely access to accurate
location information. Their increased awareness of own locations within the spatial
context would significantly reduce their chances of getting lost in buildings as well as the
associated fatalities and injuries.
It is also of critical importance for an incident commander to know the locations of the
first responders in real time, so that decision-making processes are made faster and more
informed. When an emergency happens, first response teams are sent to carry out search
and rescue operations. In most cases, searching for occupants is a manual process and
Chapter 1: Introduction
2

requires a complete inspection of all indoor spaces. Such blind search process is highly
inefficient and could be prohibited by fire, smoke or structural damage. Reducing the
time spent on searching for occupants has great potential to reduce chances of fatalities
and injuries of the trapped occupants, and it can be achieved by making the locations of
trapped occupants more visible to first responders at emergency scenes.
Despite its importance, access to the location information during emergency response
operations is far from being automated and efficient. Currently, after a size-up of an
emergency, which evaluates the severity of an incident and estimates required resources
based on visual inspections from outside a building, first response teams are sent in to the
building, usually in groups of four, to perform various tasks such as fire attack,
ventilation, and search and rescue. The deployed first responders communicate over
radios with an incident commander outside the building, who marks tasks and locations
of the deployed teams in a command post and updates this information based on vocal
reports from the deployed teams. However, it is challenging to keep this information
organized and updated, considering the ever changing situations inside the building,
especially when multiple teams use multiple radio channels to communicate. Access to
real-time location information, if made possible, would enable the incident commander to
better monitor and guide the deployed first responders. This would lead to reduction of
their chances of getting lost or trapped, and improvement of their efficiency in
performing assigned tasks. On the other hand, search for trapped occupants is usually
done in two rounds. During a primary search, first responders traverse the building,
determine a rough number and location of trapped building occupants and rescue them.
Chapter 1: Introduction
3

During a secondary search, first responders make sure all spaces are thoroughly searched,
and rescue occupants who are still trapped. Although radios (and in some cases thermal
cameras) are used to help detect the occupants at emergency scenes, the search process is
generally low-tech and blind. First responders usually have little clue of how many
occupants are trapped, where they are, and how to reach them. There is a need for an
indoor localization solution that enables the first responders to obtain real-time location
estimations of both themselves and trapped occupants during emergencies, so that they
can prioritize spaces that are more likely to have occupants when planning the search and
rescue routes, and increase their own safety during the operations.
1.2 Vision
To address the need for efficient location information of first responders and trapped
occupants during building emergency response operations, this thesis presents a
framework for real-time indoor localization. The framework locates in real time both first
responders who carry out various tasks such as fire attack, and search and rescue, and
building occupants, who are trapped during emergencies at unknown locations in
buildings. Sensor networks may not exist in buildings, or may be compromised due to
possible power outages and physical damages. Occupants may or may not have access to
devices that are typically required for localization. For the application of this localization
framework, an ad-hoc sensor network may need to be established at emergency scenes.
Localization algorithms that are capable of working with data collected by the sensor
network as well as available building information can infer locations of first responders
Chapter 1: Introduction
4

and trapped occupants. The algorithms can provide accurate location estimation, with
reasonable computational complexity and robustness.
Envisioned access to the location information has noticeable potential to improve the
efficiency of building emergency responses, improve the protection of first responders,
and reduce the chances of fatality and injuries of trapped occupants. For example, when
an emergency response team arrives at a fire emergency scene, the first responders can
rapidly set up and run the localization solution. The locations of trapped occupants can be
estimated in real time and visualized in the context of building floor plans. The location
information can be presented to an incident commander to allow for more informed
decision-making, and enhanced collaboration within first responders.

Chapter 2: Research Objectives and Questions
Motivated by the need for efficient access to location information of first responders and
trapped occupants during emergency scenes, this thesis examines the requirements of
providing indoor location information, and presents an indoor localization framework
that satisfies these requirements. The framework supports provision of location
information under various constraints imposed by emergency scenes. In the context of
this thesis, unless otherwise specified, the location information is defined as: the real-time
locations of first responders and trapped occupants in indoor environments.
The thesis has the following two objectives, which are achieved by addressing the
associated research questions:
1. Investigate the importance of location information and requirements of indoor
localization for building emergency response operations. Access to location information
is potentially valuable to the success of building emergency response operations.
However, it is among a wide range of information items that are needed by first
responders at emergency scenes, and its relative value compared to other information
items, such as weather condition and building layout, remains unclear. A close
examination of the value of location information is critical to substantiating the need for
automated access to location information. Moreover, unlike most other applications of
indoor localization, such as facilities maintenance and asset management, building
emergencies impose particular challenges and therefore may have particular requirements
Chapter 2: Research Objectives and Questions
6

that indoor localization solutions should satisfy. These requirements are yet to be
explored thus far. To achieve this objective, the following research questions are
answered in this thesis:
1.1. How valuable is the location information among various information items needed
for building emergency response operations?
In building emergency response operations, access to various information items, such
as building construction type and condition of fire, is important to improve efficiency
of all aspects of interior operations including fire attack, search and rescue, overhaul,
and ventilation. Location information is one of the valuable information items, as
knowing the locations of first responders and trapped occupants in real time enables
more informed decision-making and more efficient building emergency response
operations. To examine such value in a quantitative manner, this thesis presents the
results from an investigation of information items needed in building emergency
response operations, and assesses the value of location information among these
information items.
1.2. What are the requirements and associated metrics that should be used in the
design and evaluation of an indoor localization framework specifically for building
emergency response operations?
Unlike most environments, where indoor localization is needed, building emergency
scenes are extreme environments, imposing additional and demanding requirements
on indoor localization. For example, an indoor localization solution needs to be
Chapter 2: Research Objectives and Questions
7

quickly deployed on scene, and pre-collection of data may not be possible. A thorough
investigation of these requirements, including the metrics to measure them, is critical
to understanding nature of the problem, and to orienting design of an indoor
localization framework. These requirements and metrics should also be used in
evaluation of the indoor localization framework.
2. Examine technologies and algorithms that are capable of satisfying the above
identified requirements. A variety of technologies and algorithms have been tested by
researchers for indoor localization. Examples of tested technologies include inertial
sensors, radio frequency, and infrared. Examples of algorithms include triangulation,
fingerprinting, and sequence based localization. However, previous research mainly
focuses on the accuracy, and in some cases on the robustness, of the technologies and
algorithms. It remains unclear how these technologies and algorithms would perform
against the above requirements, and how they should be improved and integrated to
achieve satisfactory performance. To achieve this objective, the following research
questions are answered in this thesis:
2.1. What are the tradeoffs between using different indoor localization technologies?
Various technologies have been used in indoor localization in both academic research
and commercial products. Examples include the use of inertial measurement unit
(IMU), infrared, radio frequency identification (RFID), and mobile phones. Each of
these technologies has its pros and cons, and is adoptable to certain scenarios. To
examine which technologies are more capable of meeting the above-identified
Chapter 2: Research Objectives and Questions
8

requirements than the other technologies, a tradeoff analysis is carried out in the thesis,
by reviewing their uses in literature and assessing them against each of these
requirements.
2.2. What algorithm(s) should be used to infer the location of first responders and
trapped occupants?
Technologies preferred in the above assessment are used to establish a sensing
network on emergency scenes. Given the collected sensing data, a localization
algorithm is needed to compute the location of first responders and trapped occupants.
Algorithm should be compatible with the sensor data, and should satisfy the identified
indoor localization requirements.
2.3. How can the use of building information improve performance of indoor
localization?
First responders usually have access to certain building information, such as floor
plans. Building information provides knowledge of environment where the
localization is carried out. Its integration in indoor localization has potential to
improve the performance of the localization solution. It is studied in this thesis where
building information is needed in the localization algorithms, and how the integration
of building information can improve the performance of the algorithms.

Chapter 3: Review of Building Emergency
Response Procedures
When responding to building fire emergencies, fire departments across the U.S. are the
first line of defense. Upon receipt of a 911 emergency call reporting a building
emergency, a 911-dispatch center initializes an emergency response operation. The
dispatch system automatically alerts standing-by local fire stations based on pre-
determined assignment plans. Local fire stations are usually comprised of paramedic
teams and fire companies including truck companies and engine companies. Engine
companies are equipped with hoses and water so that personnel can aggressively attack
the emergency. Truck companies are like firefighter's toolbox. They carry ladders, rescue
equipment and other tools to support emergency attack activities. The type and amount of
resources dispatched for an emergency is based on type and size of the building and
severity of the incident. For example, the Los Angeles Fire Department (LAFD) dispatch
center adopts the following rule when responding to structure fire emergencies:
 Category A: Emergencies that need less than 4,500 gallon per minute (GPM) fire
flow. Buildings falling into this category include: dwellings, apartments, and hotels
that are under four stories; commercial and industrial buildings that are less than
15,000 square feet, two floors or 10,000 square feet on one floor.
The following are the minimum resources to be dispatched for such emergencies: 4
fire companies, which include at least 1 truck company. For example, the following
Chapter 3: Review of Building Emergency Response Procedures
10

resources may be dispatched for a structure fire in a three-story hotel: 3 engine
companies, 1 truck company, 1 emergency medical service captain, 2 rescue
ambulances, and 1 battalion chief.
 Category B: Emergencies that need more than 4,500 GPM fire flow. Buildings
falling into this category include: dwellings, apartments, and hotels that have four or
more stories; commercial and industrial buildings that exceed 15,000 square feet, two
floors or 10,000 square feet on one floor; schools, day nurseries that have two or
more stories; sanatoriums, homes for aged, hospitals, public assemblies, and churches.
The following are the minimum resources to be dispatched for such emergencies: 6
fire companies, which include at least 2 truck companies. For example, the following
resources may be dispatched for a structure fire in a 20,000 square feet mall: 4 engine
companies, 2 truck companies, 1 emergency medical service captain, 2 rescue
ambulances, and 1 battalion chief.
Similar categorizations are adopted by fire departments across the U.S. to determine the
type and amount of resources to dispatch for building emergencies. Local fire stations are
alerted by the dispatch center immediately and assigned a run sheet, which shows basic
information of the incident, including emergency type, time, address, and dispatched
resources. Role of the incident commander is initially assigned to the first captain
arriving on scene, and it is later transferred to higher-ranking on-scene officers. The fire
engine companies typically arrive at the scene of a fire to lead out hose lines that direct
water at the seat of the fire, and pump water in ladder trucks for large fires. Truck
Chapter 3: Review of Building Emergency Response Procedures
11

companies are responsible for searching a structure for trapped occupants, as well as
providing ventilation to help first responders deployed in the building. They are also
responsible for shutting off surrounding electric and natural gas [6].
The first duty of the incident commander upon arrival at the emergency scene is to “size
up” or make a quick appraisal of the situation. The following aspects must be assessed for
the size up [7]:
 Life hazards involved, or rescue work required, if any;
 Exposure hazards from both interior and exterior viewpoints;
 Type of building construction (for the possibilities of collapses);
 Content hazards to both occupants and firemen;
 Accessibility of the fire;
 Type and amount of fire equipment required.
After the size up, the incident commander assigns available resources to the following
tasks [7], which are repeatedly performed by shifting groups of first responders until the
situation is completely under control. Additional resources may be called in during this
process depending on the development of the emergency situation.
 Search and rescue is an operation taken to recover trapped occupants from situations
that prohibit their escape. Saving lives is always a first responder’s primary
consideration. Searching for occupants is done in two rounds: a primary search is
done at the beginning of the emergency response operation to quickly locate
occupants, who are trapped and in danger; a secondary search is done when the
Chapter 3: Review of Building Emergency Response Procedures
12

situation is under control, in order to locate trapped occupants that are not discovered
in the primary search. In both rounds, first responders need to check all of the spaces
in the building, and they mainly rely on sight, sound, and touch to discover trapped
occupants [8].
 Fire attack is an operation taken to retard or reduce the rate of burning, with an
ultimate objective to extinguish the fire. If a building is completely engulfed in flames,
the fire and heat have to be reduced before search and rescue is possible. Fire attack is
usually done with three methods: cooling or reducing temperature below the ignition
point, smothering or reducing oxygen content within the fire area below the burnable
limits, and removing fuel from the vicinity of the fire. Water is a principal fire-
extinguishing agent for fire attacks, and certain water pressure and water volume in
gallons per minute have to be ensured, depending on the type and size of the building.
 Ventilation is an operation taken to remove smoke, gases and heat from a burning
building, and to control fresh air supplies to aid the fire attack and search and rescue
operations. There are two ways to perform the ventilation, namely vertical ventilation
and cross ventilation. The former is done by making exit openings for smokes, gases
and heat on the roof. The latter is done by utilizing outside openings such as windows
and is done floor by floor.
 Salvage and overhaul is an operation taken to prevent excessive water damages, and
reassure against any possible re-ignition of fire. It is done after the fire is put out and
trapped occupants are rescued. Salvage involves covering materials stored on lower
Chapter 3: Review of Building Emergency Response Procedures
13

floors or damaged roofs with salvage covers, absorbing excessive water, and directing
water outside through doorways or other openings. Overhaul involves inspection of
concealed spaces, where fire may continue to burn unnoticed, and examination or
removal of glowing or burned materials.
 First aid is immediate and temporary care given to casualties before professional
medical personnel can treat them. First responders need to deal with the following
types of injuries: bleeding, respiratory deficiencies, shock, fractures, burns, and
wounds.
 Fire investigation is the last step taken on scene. It involves determining fire origins,
identifying fire causes, and estimating losses. If there is suspicion of arson, evidence
should be preserved.
Among the above tasks, indoor localization is directly associated with search and rescue,
where trapped occupants need to be located in the shortest time possible, and locations of
first responders going through hazardous indoor environments also need to be monitored
by the incident commander. Therefore, the task of search and rescue is the main focus of
this thesis. At the meantime, indoor location can also be of critical use to fire attack,
ventilation, and salvage and overhaul, where first responders have to work in indoor
environments before the emergencies are fully under control. When performing these
tasks, first responders need to have their locations monitored by the incident commander
to ensure their safety and improve the efficiency of decision-making and coordination.

Chapter 4: Assessment of the Value of Location
Information, and Identification of Indoor
Localization Requirements for Building
Emergency Response Operations
This chapter first examines the information items needed by first responders for
performing emergency response operations, and evaluates the respective value of these
information items, especially the location information. This chapter also examines the
requirements for performing on-scene indoor localization. The findings reported in this
chapter answer research questions 1.1 and 1.2.
4.1 Value of Location Information in Building Emergency Response
Operations
4.1.1 Card Game
A card game was designed and played with first responders to evaluate their needs for
various information items used in developing situational awareness at building fire
emergency scenes [9]. The card game created an imaginary scenario allowing an
interviewed first responder, or a “player” to assume a certain role. Information queries
and decision-making process of the player was recorded by an interviewer during the
course of a game. A card game was used because of its advantages in avoiding biases or
receipt of misleading information, which is commonly seen in face-to-face interviews. A
card game was also preferred over a written survey or a questionnaire because it
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
15

mimicked the timeline of an incident and guided a player to make sequential decisions as
the theoretic incident developed. It provided an opportunity for the participants to follow
their thought process with its open ended nature (as opposed to the structured nature of
surveys or questionnaires, where potential answers to a question are provided). It also
lessened the Hawthorne Effects, which is commonly seen in studies using surveys and
questionnaires, where respondents may improve or modify their answers when they know
the purpose and design of the study they are participating in. More importantly, unlike in
surveys or questionnaires, where respondents passively receive information to be used in
their decision making, the card game method combined and mimicked the two ways
information is delivered to the decision makers in a real-world scenario: information pull
and information push. The terms, “push” and “pull”, are originated from logistics and
supply chain management [10]. Information pull refers to cases where decision makers
initiate queries for receiving desired information [11]. Information push refers to cases
where information is delivered based on what information is assumed to be needed by
decision makers [12]. Differences between information pull and information push can
result in differences in the decision-making process, as reported in areas including
production control [13], supply chain management [14], and web based businesses [15].
As first responders may face with both ways of information delivery in decision making,
the card game integrated and represented both the pull and push methods. Information
push was represented in that a list of available information items was provided to a player
and used throughout a game. Information pull was represented in that a player had no
idea about details of a particular information item until he or she requested it, and that he
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
16

or she always had an option to request for information that was not included in the given
list.
Similar methods have been used in previous research. Muethel [16] used the card game
method to collect qualitative and quantitative data for interpreting cross-cultural settings.
Players from different cultures were asked to evaluate the importance of 16
trustworthiness-related values written on cards, and provide reasons for their evaluation.
In the construction industry, Follin and Fischbeck [17] presented eight alternatives of a
freeway design on individual cards, each showing seven features of an alternative design.
Players were asked to rank the cards based on the features, and results were used to
evaluate all alternative designs. Kiziltas et al. [18-20] adapted this method and used a
card game to differentiate the needs for historical estimation information by expert and
novice estimators in project cost estimation. The number and type of information items
requested by players were recorded and analyzed. Results revealed different behaviors of
expert and novice estimators in utilizing historical information in cost estimation. Atasoy
[21] also designed a card game in her research. Game participants were asked to build a
visualization environment they would use to monitor the project performance. The
visualization environment was built in the game by selecting different cards that indicated
different visual representations of project performance information. Results were used to
investigate how participants in construction projects would like information about project
performances to be visualized.
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
17

4.1.2 Development of Information Items List
A set of cards was prepared for the game. Each card showed a general description of an
information item that might be needed for establishing situational awareness during a
building fire emergency response operation. A list of 19 information items was developed
based on a longer list reported in [22], by removing non-fire-related information items,
and merging information items that are related. Information items in the resulting list
were organized in three categories, namely before arrival to scene, at emergency scene,
and attack and mitigation. They corresponded to three general periods of response
operations. The first period was from dispatch until arrival to an emergency scene; the
second period was from arrival until a size up was completed and methods of attack were
determined; the last period was attacking the incident until it was completely under
control [23], corresponding to tasks ”search and rescue”, “fire attack” and “ventilation”
described in chapter 3.
Information items from different categories differed in terms of when to obtain them,
how to obtain them, and how to make use of them in real-life building fire emergencies.
Information items in the “before arrival to scene” category could be obtained before an
incident occurred, and these items were generally static throughout an incident. Sources
included municipal databases, building facilities management sources, geographic
information systems (GIS), and so on. These information items helped first responders to
faster mobilize and reach an emergency scene, and better understand circumstances.
Information items in the “at emergency scene” category were related to specific features
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
18

of an incident and thus could only be obtained after an incident occurred. Examples of
sources included fire annunciator panels, smoke detectors, and security cameras. These
information items helped first responders to better understand the development of an
incident, and determine right resources and methods for incident attack. Information
items in the “attack and mitigation” category were related to search and rescue of trapped
occupants, and attack of an incident. They usually changed over the course of an incident
depending upon actions taken by first responders. These information items could help
first responders to better utilize available resources, and control an incident in a more
efficient manner. All of the information items used in the card game are listed in Table 1.
Table 1: Information Items Used in Building Emergency Response Operations
Category Number Description
Before
Arrival to
Scene
A1 Building occupancy (number and identities of occupants, based
on time of day)
A2 Building layout and site plan (building size, construction type,
floor plans)
A3 Location of water sources nearby (fire hydrants, fire
department hookups for sprinkler system, standpipes)
A4 Routing information to the building and area map of the
neighborhood of the building
A5 Contact information of building owners, managers and utility
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
19

contacts
A6 Hazards, location and identification of unusual hazards (above
ground propane tanks, gas lines, chemicals, explosives, etc.)
A7 Location of important objects (facilities, documents,
equipment) to be saved
At
Emergency
Scene
B1 Location of fire in the building, fire size, and duration
B2 Sprinklers’ status (number of location of sprinklers that have
gone off)
B3 Presence and location of occupants in the building
B4 Location and condition of smoke
B5 Warnings of structural collapse based on material type, fire
location, fire size and duration
B6 Confidence in the fire being real
Attack and
Mitigation
C1 Required water flow (gallon/minute) or foam based on fire
condition
C2 Location of available areas of refuge, staging areas
C3 Location and condition of deployed and standing-by
responding units
C4 Local weather conditions and predictions, wind direction and
velocity
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
20

C5 Locations of building entrance/exit signs
C6 Contact information of other emergency agencies

Cards were color coded to reflect the categorization and they were numbered for easy
record keeping. The cards presented high-level descriptions of information items while
details, prepared beforehand and represented with symbolic shapes, were initially hidden
from the players. During the game, when a player requested an information item by
selecting an associated card, details were released. For example, if a player selected card
“A2: Building layout and site plan”, floor plans and site plans of a building on fire would
be provided to the player. If the player then selected card “B4: Location and condition of
smoke”, symbols of smokes would be laid out on the floor plan showing the size and
location of smoke. Moreover, the player could also request any information item beyond
the provided list, and details of the requested information item were determined and
provided by an interviewer on the fly.
4.1.3 Card Game Procedures
At the beginning of a game, a player was given the following description of an
emergency scenario he or she was asked to tackle: “You are a captain. You just received
a run sheet from the dispatch center. The run sheet shows that a structural fire was
reported in a campus building, and your team is going to be the first responding unit to
arrive at the scene. You are assigned as an incident commander. Now assume you have
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
21

little background information about the emergency. Any information you need but not
given in the run sheet must be requested. You attack the incident by giving commands to
available responding units. Your goal is to put out the fire and rescue all trapped
occupants.” It needs to be noted that the scenario used in this thesis was not intended to
be representative of all types of building fire incidents. It was developed as an effort to
examine the validity of this approach in understanding the information needs of first
responders. The card game was developed for an office building in collaboration with the
first responders from the Los Angeles Fire Department. The scenario, however, could be
adapted to other types of buildings by modifying the incident size and type, and the list of
information items. The information needs might vary based on the type of buildings and
the severity of fires, and such variations should be examined in follow-up studies.
The game is divided into multiple rounds, and each round is divided into three phases.
The number of rounds is determined based on decisions a player made and rules
described below. In phase one of each round, a player could select one card from each of
the three categories (i.e. before arrival to scene, at emergency scene, and attack and
mitigation). The player can also select no card, if he or she believes no further
information item from a category is needed. Whenever a card is selected, an interviewer
releases associated details as described above. In phase two, a player can request an
update of information items s/he possesses. For example, if a player requests an update
on “B4: Location and condition of smoke”, an interviewer will tell him/her whether
smoke is increased or decreased after his/her commands given at the end of the last round
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
22

are carried out. In phase three, a player is asked to describe his/her assessment of current
situations, and to give new commands.
The game continues until the goal, i.e., “put out the fire and rescue all trapped occupants”
as given in the scenario description, is achieved. Development of an incident is
determined by the following rules (which are not disclosed to players): There are four
units of fire in four rooms along both sides of a corridor on one floor, and four units of
smoke around the burning area. Ten occupants are trapped in two rooms. The fire and
smoke are measured in “unit” for simplicity purposes. If responding units are assigned to
attack fire, both fire and smoke are reduced by one unit, except when a player used
extensive resources (i.e., more than three responding units), both fire and smoke are then
reduced by two units. If no responding unit is assigned to fire attack in a round, fire and
smoke remain the same. On the other hand, if a search and rescue task is carried out for
two consecutive rounds, occupants in one room are saved.
After a player achieves the goal, he or she is asked to review all of the cards prepared for
the game and their details, and rank the cards within each of the three categories based on
the cards’ overall value in establishing situational awareness that supports building fire
emergency response operations.
4.1.4 Card Game Implementation
From May 10, 2012 to June 17, 2012, the card game was played with 29 first responders
from 8 local fire stations, 2 firefighter-training centers, a 911-dispatch center and a
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
23

university campus safety department in the greater Los Angeles area. Players included 13
captains (including 2 battalion chiefs), 13 firefighters and paramedics, and 3 engineers.
Among them, 7 were the California Urban Search and Rescue (US&R) team members
from 2 US&R task forces. The US&R team members are all specially trained firefighters
responding to localization, extrication, and initial medical stabilization of victims trapped
in confined spaces due to emergencies such as natural disasters, structural fires and
collapses, transportation accidents, mines and collapsed trenches. Players’ years of
experience varied between 4-30 years, with an average of 14.4 years.
4.1.5 Analysis of Card Game Results
The importance of an information item was assessed by evaluating three aspects [9]: (1)
the order it was requested. Given limited time for information collection and decision-
making, incident commanders tended to first request information items that they believed
were the most critical. The earlier an information item was requested, the more critical it
was to building fire emergency response operations; (2) the frequency that its update was
requested. Some information items, such as location of occupants in the building, were
dynamic, thus deserved continuous attention of incident commanders as an emergency
evolved; (3) its overall value ranked by the players at the end of the game.
One-sample t-test was used to estimate the average order in which information items
were requested. For each card X and each possible order k (minimum = 1, maximum =
the number of information items in X’s category), the following null hypothesis was
tested: “X is likely to be the k-th card requested in this category”. Results are summarized
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
24

in Table 2. At a 95% confidence level, card “A4: Routing information to the building and
area map of the neighborhood of the building” was requested first in the “before arrival to
scene” category, with the highest level of agreement among players. The level of
agreement was measured by the covariance of orders at which players selected an
information item in the game. A smaller covariance indicated a higher level of agreement.
A4 was followed by “A1: Building occupancy (number and identities of occupants, based
on time of day)” and “A3: Location of water sources nearby (fire hydrants, fire
department hookups for sprinkler system, standpipes)”. The card “A5: Contact
information of building owners, managers and utility contacts” was lastly requested in
this category, with the lowest level of agreement among all players. In the “at emergency
scene” category, “B1: Location, size, and duration of fire in the building”, “B3: Presence
and location of occupants in the building” and “B4: Location and condition of smoke”
were the top three cards requested, with similar levels of agreement among players, while
“B2: Sprinklers’ status (number of location of sprinklers that have gone off)” was the
least requested card. It was interesting that 72.4% of the players did not select “B6:
Confidence in the fire being real”; however, for those who did, they all requested it at the
very beginning. In the “attack and mitigation” category, there was a high level of
agreement that “C3: Location and condition of deployed and standing-by responding
units” should be first requested, followed by “C2: Location of available areas of refuge
and staging areas” and “C1: Required water flow (gallon/minute) or foam based on fire
condition”. It needs to be noted that, in C3, the conditions of responding units included a
range of factors such as first responders’ physiological status and health, and signs for
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
25

overexertion. These factors could determine the capability of first responders in
performing their assigned tasks and avoiding potential risks, and hence should be
monitored closely. The card “C6: Contact information of other emergency agencies” was
requested the last by players who selected it.
Frequency analysis, as presented in Table 2, showed that “A6: Hazards, location and
identification of unusual hazards (above ground propane tanks, gas lines, chemicals,
explosives, etc.)” was the most frequently updated card in the “before arrival to scene”
category. It was on average updated 1.33 times per player, indicating its dynamic nature
and high relevance to the success of response operations. “A5: Contact information of
building owners, managers and utility contacts” was never updated by any player, due to
its static nature. In the “at emergency scene” category, update frequency increased
significantly. Over 86% of the players requested an update on “B3: Presence and location
of occupants in the building” and “B1: Location of fire in the building, fire size, and
duration”. They were updated 3.3 and 3.2 times per player, respectively. “B6: Confidence
in the fire being real” was not requested to be updated by any player. In the “attack and
mitigation” category, “C3: Location and condition of deployed and standing-by
responding units” was the only information item that was frequently updated, with an
average of 2.21 updates per player. “C4: Local weather conditions and predictions, wind
direction and velocity” and “C5: Locations of building entrance/exit signs” were never
updated by any player. In addition, it needs to be noted that, compared to information
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
26

items in other categories, information items in the “at emergency scene” category were
updated on average five times more frequently.
Similar to the order analysis, one-sample t-test was used to estimate the overall value of
an information item assessed by players at the end of the game. For each card X and each
possible importance rank k (minimum = 1, maximum = the number of information items
in X’s category), the following null hypothesis was tested: “X represents the k-th
valuable information item in this category”. At a 95% confidence level, “A4: Routing
information to the building and area map of the neighborhood of the building” and “A1:
Building occupancy (number and identities of occupants, based on time of day)” were
considered the first and second valuable information items in the “before arrival to scene”
category, with close ratings by players and moderate levels of agreement. “B1: Location
of fire in the building, fire size, and duration” and “B3: Presence and location of
occupants in the building” were the first and second valuable information items in the “at
emergency scene” category, with high levels of agreement among players. “B2:
Sprinklers’ status (number of location of sprinklers that have gone off)” was the least
valuable in this category. In the “attack and mitigation” category, “C3: Location and
condition of deployed and standing-by responding units” was considered the most
valuable and “C6: Contact information of other emergency agencies” was considered the
least valuable. The results are summarized in Table 2.
.
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
27

Table 2: Summary of Card Game Results
Order Frequency Importance
Before
Arrival to
Scene
1
st
A4: Routing
information to the
building and area
map of the
neighborhood of the
building
A6: Hazards,
location and
identification of
unusual hazards
(above ground
propane tanks, gas
lines, chemicals,
explosives, etc.)
A4: Routing
information to the
building and area
map of the
neighborhood of the
building
2
nd
A1: Building
occupancy (number
and identities of
occupants, based on
time of day)
A3: Location of
water sources nearby
(fire hydrants, fire
department hookups
for sprinkler system,
standpipes)
A1: Building
occupancy (number
and identities of
occupants, based on
time of day)
3
rd
A3: Location of
water sources nearby
(fire hydrants, fire
department hookups
for sprinkler system,
standpipes)
A7: Location of
important objects
(facilities,
documents,
equipment) to be
saved
A3: Location of
water sources nearby
(fire hydrants, fire
department hookups
for sprinkler system,
standpipes)
At
Emergency
1
st
B1: Location, size,
and duration of fire
B3: Presence and
location of occupants
B1: Location of fire
in the building, fire
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
28

Scene in the building in the building size, and duration
2
nd
B3: Presence and
location of occupants
in the building
B1: Location of fire
in the building, fire
size, and duration
B3: Presence and
location of occupants
in the building
3
rd
B4: Location and
condition of smoke
B4: Location and
condition of smoke
B4: Location and
condition of smoke
Attack and
Mitigation
1
st
C3: Location and
condition of
deployed and
standing-by
responding units
C3: Location and
condition of
deployed and
standing-by
responding units
C3: Location and
condition of
deployed and
standing-by
responding units
2
nd
C2: Location of
available areas of
refuge and staging
areas
C2: Location of
available areas of
refuge, staging areas
C1: Required water
flow (gallon/minute)
or foam based on fire
condition
3
rd
C1: Required water
flow (gallon/minute)
or foam based on fire
condition
C1: Required water
flow (gallon/minute)
or foam based on fire
condition
C2: Location of
available areas of
refuge, staging areas

The impact of players’ profile on the use and evaluation of information items was
examined using a Friedman test. Specifically, the impact of years of experience, job title,
and US&R team membership on the order and overall value of information items were
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
29

analyzed. These three factors were chosen for analysis because intuitively the norms of
players could change as their experience grew, and their information needs could vary
based on their positions and assigned tasks. Besides, additional training that some players
received as US&R team members could reshape their information needs. The impact of
these factors on update frequency was excluded from the analysis due to incomplete data
set, as not every information item was updated by every player.
After dividing all players into groups based on their profile, a Friedman test was used to
examine whether the order and the overall value of each information item were consistent
among different groups. At a 95% confidence level, the following conclusions could be
drawn based on the Friedman test results: years of experience had no statistical impact on
any information item except for “B4: Location and condition of smoke”. Job title or
US&R membership had no statistical impact on any information item. Lastly, it is worth
pointing out that few players requested any information item beyond the given list, which
suggested that the information items list used in this thesis was comprehensive and
sufficed players’ needs.
A summary about location-related information items, in particular, emphasized that
location of trapped occupants represented by “B3: Presence and location of occupants in
the building” was the most frequently requested and one of the earliest-needed
information items in the “at emergency scene” category. It was also ranked as the second
most important information item in this category. Such high frequency and sequence of
request and high importance indicated that the location of trapped occupants was
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
30

significantly valuable in emergency response operations and deserved continuous
attention. This finding was also a strong demonstration of an agreement among all
players that saving occupants was one of the two foremost crucial tasks in emergency
response operations. The other task was putting out the fire, partly represented by “B1:
location, size, and duration of fire in the building”.
Moreover, location of first responders, represented by “C3: location and condition of
deployed and standing-by responding units”, was also considered a valuable information
item. It was the most frequently requested and the first needed information item in the
“attack and mitigation” category. It was also ranked by the players as the most important
information item in this category. The results indicated that access to the location of first
responders played a crucial role in supporting the decision-making and implementation of
emergency response operations. Continuous monitoring of the real-time location of
deployed first responders was therefore highly valued and desired.
4.2 Indoor Localization Requirements for Building Emergency
Response Operations
Most of the existing indoor localization solutions designed for building emergency
response operations are highlighted by either their high accuracy or their independence
from existing infrastructure. However, it remains unclear what level of accuracy is
sufficient to support emergency response operations. Although a higher accuracy is
always desired, it may require a sophisticated sensing network or additional prior data
input, or cause other issues that are undesired by first responders. Independence from
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
31

existing infrastructure is desired because it increases the robustness of a solution.
However, robustness of a solution is also impacted by various other factors, such as
reliance on building power supplies, and resistance to heat, water and smoke. These
challenges are all imposed by emergency scenes and require further examination.
Moreover, prior research discussed little about requirements other than accuracy and
robustness, whereas the performance of an indoor localization solution in terms of e.g.
cost, ease of deployment, and computational speed may be important to emergency
response operations. All of these questions point to the need for a thorough examination
of indoor localization requirements for emergency response operation, including their
metrics. Such an examination is critical to understanding the nature of the localization
problem, and to orienting the design of indoor localization solutions. These requirements
and metrics should also be used in evaluation of any indoor localization solution that is
intended to support building emergency response operations.
4.2.1 Survey Design
In order to examine the indoor localization requirements for building fire emergency
response operations, a survey was designed by the authors and distributed to first
responders in the United States [24]. A list of eleven possible requirements was
developed and used in the survey. The list was generated based on extensive discussions
with the first responders from the LAFD, and went through several rounds of revisions
based on the feedback received from the LAFD. The survey took into account both the
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
32

technical requirements of indoor localization and practical requirements of implementing
a solution at emergency scenes. These possible requirements are summarized in Table 3.
After providing their profile information, including their organizations, years of
experience and email addresses, the respondents were first asked to choose five of the
requirements that they thought were important for locating first responders and trapped
occupants at building fire scenes. The rest of the survey focused on understanding the
details, especially metrics, for evaluating each of these possible requirements.
The survey was hosted by an online surveying tool. This web based survey tool kept a
record of the IP addresses of the computers, from which the surveys were completed, and
assigned an ID to each response. By checking the IP addresses, response IDs, and
respondents’ profiles, responses that were duplicates or not responded by the targeted
first responders were excluded from the analysis.
The survey was open for 14 weeks from August 10 to November 16, 2012. It could be
accessed at this link: https://uscviterbi.qualtrics.com/SE/?SID=SV_1TEtF7yTbFohcfr.
First responders were invited from all 50 states of the U.S. to participate in the survey.
Contact information of first responders was acquired by manually searching the websites.
The qualified websites included official websites of the state fire marshal offices, city fire
departments and local fire stations, and websites of organizations and communities
related to building fire emergency response operations, such as the National Fire
Protection Association (NFPA). The qualified entries of contact information included
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
33

those that contained a line officer's full name, affiliation, job responsibility and email
address. Survey invitations were successfully delivered via email to 1151 first responders
across the country. The survey invitation recipients were all line officers with various
rankings. Civil officers (e.g. presidents, board members, secretaries) who do not perform
on-site jobs were not invited. A total of 283 responses were received. After filtering the
incomplete, duplicate or invalid responses, 197 valid responses were used for analysis,
which supported a ± 6.8% confidence interval of the results at a 95% confidence level. It
needs to be noted that the survey sampling was limited by the accessibility to online
contact information. Fire chiefs and captains comprised a larger proportion in the sample
than their true proportion in the real world, and first responders, whose affiliations were
not listed on public websites, were not sent a copy of the survey. This survey was not an
attempt to accurately describe the opinions of first responders in the most representative
way. Rather, it was an attempt to sample a set of beliefs that depended on expertise and
training levels of the participants.
4.2.2 Survey Analysis
Among all of the survey respondents, the years of experience varied between 2 to 44
years, with an average of 25.7 years. The composition of the respondents is shown in
Figure 1. Chiefs/commissioners comprised the largest group (18.78%), followed by the
battalion chiefs or district chiefs (14.72%), captains (14.72%), and assistant
chiefs/commissioners or deputy assistant chiefs/commissioners (11.68%). Respondents
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
34

who reported “others” included fire marshals, assistant fire marshals, fire investigators,
and emergency coordinators.

Figure 1: Job titles of the survey respondents
Based on the survey results, all possible requirements were organized in a descending
order according to the response rates (Table 3). The survey results showed that the most
important requirements were: accuracy, ease of on-scene deployment, robustness
(resistance to heat, water and other physical damages), computational speed (speed of
calculating and presenting location information), and size and weight of devices. All of
these five requirements were considered important by more than half of the responders,
which was remarkably higher than the response rate of all other requirements (13.7% to
38.7%).

Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
35

Table 3: Importance of indoor localization requirements
Rank Requirement Percentage of respondents who
considered this requirement
important
1 Accuracy of location information 90.4%
2 Ease of deploying the solution on
scene 83.8%
3 Resistance to heat, water and other
physical damages 67.0%
4 Speed of calculating and presenting
location information 66.0%
5 Size and weight of devices attached
to first responders and occupants 58.9%
6 Purchase and maintenance costs 38.7%
7 Independence from building
infrastructure (e.g. building
components and installed
equipment) and building power
supplies 22.8%
8 Independence from prior data
collection (e.g. building layouts
obtained by access building data,
14.2%
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
36

Further details of these requirements were examined in the survey. Based on the survey
results, room-level accuracy was the most useful level of accuracy desired by first
responders, rated at 4.21 out of 5. It was followed by floor-level accuracy, rated at 4.09.
Meter-level accuracy was less desired, rated at 3.59. Its low desirability was probably due
to that such high accuracy did not provide further information beyond the room-level
accuracy, but added to the difficulty in interpreting numerical location information. Sub-
meter level of accuracy was the least wanted, rated at 2.15 by the respondents.
Ease of on-scene deployment was measured by time spent on deploying a solution on
scene. The responders reported that a maximum of 2.25 m, or 135 s, was allowed to be
spent on on-scene deployment.
and radio features obtained by
surveying a building)
9 Scalability of the solution to cover
large numbers of people 14.2%
10 Ease of assembling the solution
before dispatch 14.2%
11 Independence from on-scene data
input (e.g. known coordinates of a
few locations inputted by first
responders) 13.7%
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
37

Robustness of a solution was measured by its resistance to damages (heat, water and
physical damages). Emergency scenes impose various challenges to the robustness.
According to survey results, some challenges cause major concerns, including (1) heat; (2)
water; (3) smoke; and (4) physical damages. Therefore, a localization solution is desired
to have devices and algorithms that are robust against these challenges. Other challenges
cause less concerns, including (5) loss of building infrastructure, such as routers and
artificial lighting, that an indoor localization solution may rely on to operate; (6) loss of
building power supplies; and (7) unavailability of pre-collected data, including building
layouts and radio signal propagation characteristics used as the input in location
computation.
As for devices attached to or carried by first responders, survey respondents preferred
devices to be equal to or smaller than 107.34 cm
3
, and equal to or lighter than 1.16 kg. In
addition, 54.82% responders reported that devices should be fit into their coats or
backpacks.
In terms of computational speed, an appropriate time reported by responders for data
processing/location computation varied from 5 to 180 s, with an average of 40.34 s. It
included the time spent on collecting data from a sensing network, implementing a
localization algorithm to compute targets’ locations, and presenting localization results to
incident commanders.
Chapter 4: Assessment of the Value of Location Information, and Identification of Indoor
Localization Requirements for Building Emergency Response Operations
38

4.3 Conclusions
The value of access to location information to the success of building emergency
response operations is examined in this chapter. The results from the card game highlight
the necessity to provide first responders with effective access to location information
during their operations, which is expected to improve the efficiency of finding occupants
trapped in burning buildings, and reducing the chances of fatalities and injuries of both
trapped occupants and first responders. The results of a nationwide survey, which
examined the requirements for indoor localization at building emergency scenes, are also
reported in this chapter. The survey results show that an effective indoor localization
solution should have satisfactory performance based on the following criteria: accuracy,
ease of on-scene deployment, robustness against various physical damages,
computational speed, and size and weight of devices. These findings lay the basis of the
development of an indoor localization framework that is explained in the following
chapters.

Chapter 5: Review and Evaluation of Indoor
Localization Technologies
This chapter reviews existing solutions to indoor localization that are intended for
building emergency response operations in specific and for various other applications in
general, and assesses the performance of a number of technologies that have been used in
indoor localization.
5.1 Review of Indoor Localization for Building Emergency Response
Operations
Regardless of the high value of indoor location information for building emergency
response operations, there is no single indoor localization solution that has been widely
adopted by first responders. However, there are a few solutions proposed. For example,
ultra wide band (UWB), a type of radio frequency (RF) technology, was used in several
solutions. Akcan and Evrendilek [25] proposed a system that was built on the UWB
technology. Two radios were used on every node, which enabled directional localization
in static networks. Reported accuracy through simulations was up to 6 m, depending on
the node density. Another UWB based system was proposed by Harmer et al [26]. The
system used a time difference of arrival (TDOA) based algorithm for 3D location
estimation. Reported accuracy in tests varied between 1 m and 2 m. The system required
deployment of a large amount of sensors, and could not locate building occupants that
had no access to special mobile units, which were part of the sensing network and
Chapter 5: Review and Evaluation of Indoor Localization Technologies
40

required for localization. Rantakokko et al. [27] reviewed existing technologies for
indoor localization, and proposed a system that integrated foot-mounted inertial sensors
and UWB sensors to support first responders. Field tests reported accuracy between 1 and
4 m. The system suffered from heading drifts depending on the travel distance.
Solutions that use other types of technologies also exist. Chandra-Sekaran et al [28,29]
proposed a system to locate patients during emergencies in both outdoor and indoor
environments. Radio nodes were attached to doctors and patients. Both Monte Carlo and
Unscented Kalman Filter techniques were used for location estimation. The system was
evaluated through simulations, and reported an accuracy that varied between 5 and 10 m.
Duckworth et al. [30] proposed an indoor localization system for emergency response
operations that required no existing infrastructure or pre-characterization of the area of
operation. The system relied on an ad-hoc network built on transmitters carried by both
first responders in a building and vehicles outside the building. In a series of publications
[30-33], the authors reported up to sub-meter accuracy. The system required a
considerable amount of infrastructure that could require an excessive effort in on-scene
deployment and maintenance. Khoury and Kamat [34] tested a commercial wireless local
area network (WLAN) based localization system in an emergency response training
facility used by urban search and rescue professionals. Accuracy between 1.5 and 2 m
was reported, and results were visualized on the 2D layout of the facility. Guerrieri et al.
[35] proposed a framework for locating first responders with an RFID based localization
system. Unlike similar systems that use nodes to track occupants and placed fix readers in
known locations, their system fixed nodes in known locations and attached readers to
Chapter 5: Review and Evaluation of Indoor Localization Technologies
41

tracked occupants. No prototype was built or evaluated. Ruppel et al. [36] proposed a
framework to support rescuers in emergency or during daily maintenance work. The
system utilized UWB in halls, WLAN in office areas, and RFID in infrastructure-free
rooms such as cellar rooms or underground garages. It also integrated building
information modeling (BIM) technology to provide rescuers with building information of
their immediate surroundings. Preliminary tests reported an accuracy of up to 3 m. Kaya
et al. [37] used a backward ray-tracing algorithm to analyze angle of arrival (AOA), time
of arrival (TOA) and signal power for locating first responders, who wore beacons. Using
multiple receivers, they were able to cover 80% of a building and achieve an accuracy of
within 10 m.
There are also a few commercial solutions available in the market. Stemming from
research sponsored by the Department of Homeland Security, SPIE’s [38] geospatial
location accountability and navigation system for emergency responders (GLANSER)
used a combination of various technologies including a global positioning system (GPS),
an inertial measurement unit (IMU), an UWB, a Doppler radar, as well as a
magnetometer, a compass, a pedometer, and an altimeter inside a wearable electronic unit.
The details of the algorithm were not disclosed, but an accuracy of 3 m was claimed to be
achievable in field tests. Exit Technologies [39] provided another solution that used
handheld devices operating at low-frequency radios. A distressed first responder
attempting reorientation or self-rescue could send out signals with a handheld device. The
signals could then guide other first responders to the transmitting device. No details of
the algorithm or its accuracy were disclosed.
Chapter 5: Review and Evaluation of Indoor Localization Technologies
42

5.2 Review of General Indoor Localization Solutions
Given the limited indoor localization solutions designated for building emergency
response operations, this subchapter provides a review of general indoor localization
solutions available in the literature, in order to have a thorough examination of
technologies and algorithms that are applicable for indoor localization and have potential
to be used at building emergency scenes.
Location information is crucial to a variety of standard and personalized applications in a
wide range of industries such as transportation, manufacturing, logistics, and healthcare,
and it is the basis for delivery of personalized and location based services [40]. Examples
include applications that show drivers their vicinity and guide them to their destinations
[41], applications that enable a user to search for published services within his/her
immediate vicinity [42], and applications that monitor a user's indoor location, and adjust
the music played in rooms based on user preferences [43]. While GPS technology has
well met the need for outdoor location sensing, indoor localization has generally
remained a challenge and attracted considerable attention in both the industry and the
academia. Various technologies have been proposed and tested for indoor localization.
These technologies include: inertial navigation system (INS), assisted GPS (AGPS),
infrared, and RF technologies (such as UWB and RFID). A holistic review of these
indoor localization technologies is carried out in this subchapter.
Chapter 5: Review and Evaluation of Indoor Localization Technologies
43

5.2.1 Review of INS Based Solutions
INS is built on inertial sensors that measure acceleration, velocity, orientation and
gravitational forces. Recent technological development has also led to the use of micro
electromechanical systems (MEMS), which make INS more affordable and portable.
Indoor localization with INS is usually done with a pedestrian dead reckoning (PDR)
technique, which estimates a person’s current position by advancing a previously
determined position based upon an estimated speed over elapsed time along an estimated
direction. Errors of locations estimated using the PDR technique could be below 5% of
total travelled distance [44].
A major constraint of the PDR technique is that it assumes a constant step length, which
cannot be met at all times and can lead to significant errors. To overcome this limitation,
a zero velocity update (ZUPT) method is widely used [45,46]. If mounted on a foot,
inertial sensors should have zero relative velocity to the ground when the foot is in
contact with the ground during a stride, which is called a still-phase. Therefore, resetting
the velocity to zero when a still-phase is detected can reduce accumulated errors from
accelerometer output. However, this calibration method becomes invalid if people run,
when both legs can leave the ground. To alleviate the issue of people’s behaviors,
Yamanaka et al. [47] built a solution that integrated inertial sensors and electromagnetic
sensors. The solution estimated a moving distance in the period when a person’s both
legs did not ground by estimating the velocity of waist. Reported accuracy was below 5%
of the travelled distance.
Chapter 5: Review and Evaluation of Indoor Localization Technologies
44

Despite these calibration methods, errors of INS accumulate over time, and can increase
to unacceptably large as travelled distances increase. To ensure a stable accuracy, INS is
usually calibrated periodically with data collected using other technologies, such as GPS
[48,49], odometer and active beacons [50], RFID [51], cameras [52] and laser scanners
[53]. Accuracies of these calibrated solutions could be improved to up to a few
centimeters irrespective of travelled distance. While integration of other technologies has
led to increased accuracies, it also has certain downsides: it requires additional
infrastructure, and makes a solution no longer self-contained. To overcome these
downsides while ensuring good accuracies, researchers have explored the integration of
building information. This idea originated from car navigation, where a car is assumed to
be travelling on a road and non-holonomic constraint is valid [54]. Particle filters [55] or
map mapping heuristics [56] can be used to constrain estimated locations, based on
integrated building information. A person would be repositioned if his/her trajectories
estimated with IMU data do not fit into the building geometries [57]. These building
information-enabled solutions could achieve accuracies of within 0.73 m 95% of the time
in a three-story building.
5.2.2 Review of AGPS
GPS signals are usually weak or lost in indoor environments, making it impossible to
perform accurate indoor localization with conventional GPS devices. However, the
success of outdoor GPS localization has motivated research on modifying GPS to make it
applicable in indoor environments. This has led to the introduction of assisted GPS, or
AGPS. AGPS receives additional data from networks, such as satellite orbital data and
Chapter 5: Review and Evaluation of Indoor Localization Technologies
45

precision time, so that it can locate users with faster speed and higher accuracy than
conventional GPS in indoor localization [58]. High sensitivity GPS (HSGPS) receivers
are sometimes used together with AGPS. Held by users, these receivers provide extra
processing power that helps boost weak signals to a point where they can be used to
provide a positioning or timing solution. Certain signal acquisition algorithms have also
been designed to improve the signal acquisition capacity of APGS [59,60].
Digglen et al. [58] proposed an AGPS solution a decade ago, which could utilize satellite
orbit information to achieve faster cold-starts, and perform massive parallel to enable
robust location computation. An accuracy of 17 m was achieved in a test done in a
shopping mall. Bayrak et al. [61] made an early effort in developing an AGPS solution in
which messages from assistance antennae were sent via IP data connections instead of
pre-defined cellular network. System communication capacity exhaustion was avoided
this way. Field tests showed an improvement of accuracy over conventional GPS by 34 -
42 m in urban indoor environment. Ozsoy et al. [62] used GPS repeaters that consisted of
directional antennae for receiving a non-overlapping set of GPS satellites. They also used
power amplifiers to compensate for cable and antenna losses, and transmitting antennae
for re-radiating amplified GPS signals. Field tests showed an accuracy of 5 m in 2D
indoor localization. Anwar et al. [63] proposed to use a combination of AGPS and
HSGPS to locate indoor users. Antennae of AGPS were installed on a roof, which
received satellite signals and sent them to users. Users could use HSGPS to receive
signals from AGPS, and use geometric relations based on signals’ travel time to infer the
users’ distances to satellites, and then to calculate the users’ locations. Simulations
Chapter 5: Review and Evaluation of Indoor Localization Technologies
46

yielded an accuracy of within 6 m. They also concluded that an extended Kalman filter
yielded better results than a nonlinear least square method in fusing data [64].
5.2.3 Review of Infrared Based Solutions
A passive infrared (PIR) sensor is an electronic sensor that measures infrared light
radiating from objects in its field of view. When used in indoor localization, the intensity
of infrared light is used to estimate a distance between a sensor and a radiator, and the
distance is then used to locate a target with triangulation or proximity estimation [65,66].
In an indoor localization solution for robots proposed by Jijikata et al. [67], a number of
infrared light emitting diodes (LEDs) were deployed in known locations. Robots detected
the LEDs to determine their own locations and orientations. An accuracy of about 4 m
was reported in simulation. A similar solution was proposed in [68] with a reported
centimeter level accuracy in simulation in a 1.5 × 2 m
2
area. Luo et al. [69] introduced a
dynamic triangulation algorithm, which fused data from PIR sensors and RF sensors.
Simulations reported a mean accuracy of 0.73 m. PIR sensors can also emit infrared light
which, when hits an object, is bounced back and detected by the PIR sensor. The round
trip time can be used to estimate a distance of the object from the sensor [70]. Petrellis et
al. [71] used PIR sensors in a different way. They proved that the success of pattern
recognition of signals sent by PIR transmitters could be analyzed to indicate the location
and orientation of a target relative to location-known transmitters. Simulation reported an
accuracy of 10 cm in a 5 m
2
area. In addition, researchers have used passive thermal
infrared sensors for indoor localization. Kemper and Linde [72] proposed to use passive
thermal infrared sensors to detect thermal radiation of humans, and then use triangulation
Chapter 5: Review and Evaluation of Indoor Localization Technologies
47

to locate persons in a room. In simulations in a 22.5 m
2
area, an accuracy of 20 cm was
reported. A similar solution was proposed in [73,74] that reported a maximum error of 68
cm in a similar scale of simulation.
5.2.4 Review of RF Based Solutions
There are a number of RF technologies that have been applied to indoor localization.
Ordered from low frequency to high frequency, these technologies include RFID (125
kHz – 956 MHz), WLAN/WiFi and Zigbee (2.4 GHz), and UWB (3.1 GHz – 10.6 GHz).
5.2.4.1 RFID Based Solutions
A typical RFID system consists of two components, namely a reader and a tag. A tag
contains a microchip and an internal antenna. Attached to an object, a tag stores a specific
ID and other object-related data, and sends the data to a reader upon its request.
Early RFID based indoor localization solutions [75] had low accuracy, mainly due to the
reliance on an empirical radio signal propagation function that was subject to multipath
effects, a phenomenon where signals reach receivers by more than one path. Later
research used a room with few partition walls, laid out sponges on walls to absorb
magnetic wave reflection, and reported increased accuracies [76]. Luo et al. [77] added to
the literature by testing effectiveness of four different algorithms, including three
variances of triangulation. Field tests done in a building and a construction site reported
an accuracy of up to 1.2 m.
Chapter 5: Review and Evaluation of Indoor Localization Technologies
48

LANDMARC [78] was a fundamental solution. It used a proximity based algorithm.
Field tests done within an area of 4 × 10 m
2
reported an accuracy of within 2 m with 90%
probability. In order to increase the accuracy of LANDMARC, a number of research
projects followed, by improving algorithm designs [40,79-82], extending algorithm
capabilities [83-86], calibrating estimated locations with additional computation [87,88],
and optimizing deployment of reference tags [89,90]. Further recent efforts in RFID
based indoor localization have been done. These efforts experimented with localization
algorithms other than triangulation and proximity [91], integrated RFID technology
together with other technologies [92], and tested solutions in zone or building-scale
deployments [93].
5.2.4.2 WiFi/WLAN Based Solutions
WiFi, also referred to as WLAN, is a widely used technology that allows wireless data
exchange between electronic devices over radio frequency in a 2.4 GHz band. Routers
are used in a WiFi network, serving as access points (APs) to connect devices inside the
network to external networks (e.g. the Internet).
Existing WiFi based localization is mostly done with fingerprinting, where two phases
are involved: a training phase that generates a radio map comprised of radio signal
fingerprints, and an application phase that identifies the best matching fingerprints as
users’ estimated locations.
For example, Grossman et al. [94] evaluated three methods for identifying the best
matching fingerprints. These methods included: minimum Euclidean distance,
Chapter 5: Review and Evaluation of Indoor Localization Technologies
49

intersections of received signal strength indication (RSSI)-Isolines, and a stochastic
model based on Bayes’ theorem. Field testes reported accuracies between 2 and 3 m for
all three methods. Gansemer et al. [95] extended the first two methods to 3D, and
reported a recognition rate of over 93%. Fang and Lin [96,97] proposed to transform
RSSI values into principal components such that information of all APs can be more
efficiently utilized. Field tests showed that accuracy could be improved by 33.75% and
online computational load could be reduced by 40% simultaneously. Schmitzberger [98]
shifted computational load away from mobile devices towards a background localization
infrastructure. The resulting solution was autonomous, and yielded an accuracy of around
4 m with 80% probability.
A major limitation of the fingerprinting algorithm is the training phase being time- and
labor-consuming. Part of the fingerprints can be interpolated to alleviate this problem, at
the cost of an increase of error distances by 18% [99]. Corte-Valiente [100] made a step
further, proposing to generate an entire radio-map by interpolation. RSSI values of
reference locations were estimated using a radio signal propagation model that took into
account signal attenuation through walls by using a 2D ray tracing method. Accuracy
reported in simulation was around 0.2 m. The approach was extended to 3D in [101].
Wang et al. [102] proposed an approach to dynamically generate radio maps to avoid
their reconstruction over time. The relationship between RSSI values of a few calibration
points, whose RSSI values were periodically updated, and RSSI values of reference
points was established using artificial neural network. Other methods were also proposed
in literature to improve the performance of WiFi based indoor localization solutions.
Chapter 5: Review and Evaluation of Indoor Localization Technologies
50

These included mitigating multipath effects [103], improving identification of matching
fingerprints [104], and optimizing number and location of APs [105,106].
5.2.4.3 UWB Based Solutions
UWB is a technology for transmitting information spread over a bandwidth larger than
500 MHz. When used in indoor localization, it usually requires a denser deployment of
equipment and yields a higher accuracy than other RF technologies. RSSI, TOA and
TDOA are the most widely used approaches for UWB based localization. For example,
Bocquet et al. [107] used a simple input multiple output approach using TDOA readings,
which resulted in an accuracy of 2 cm in simulation. Success of TOA or TDOA
algorithms depends on the quality of TOA or TDOA readings and ultimately the accuracy
of synchronization between beacons. Ye et al. [108] quantified localization errors using a
TOA approach, and proposed a new leading-edge algorithm to eliminate timing errors. A
centimeter level accuracy was reported in their simulation. Xiong et al. [109] proposed a
revised TOA algorithm. In their solution, a receiver sent a signal back to a transmitter
using a different carrier frequency. Difference of trip time between the original and the
feedback signals was used for distance estimation that formed the basis of triangulation.
A centimeter level accuracy was reported in their simulation.
Like other RF technologies, UWB also suffers from multipath effects. Leucken and
Wittneben [110] proposed a UWB based system that used a large signaling bandwidth to
reduce the multipath effects, and a low sampling rate to increase ease of implementation.
An accuracy of up to 20 cm was reported in simulation. Meissner el al. [111] reported
Chapter 5: Review and Evaluation of Indoor Localization Technologies
51

their work on modeling deterministic part of multipath channels by using virtual signal
sources and analyzing impacts of building geometries. About 90% of UWB channel
impulse responses could be explained by their approach in terms of energy capture. Other
improvements have also been made to UWB based solutions. Li et al. [112] proposed a
solution that could optimize placement of UWB receivers, and detect and eliminate
ambiguous results. An accuracy of 10 cm was reported in simulation in a 125 m
3
cubic
space. Ingram et al. [113] proposed an integration of UWB with GPS. No results were
obtained and reported at the time of publication.
In addition, UWB can be used in confined small spaces to provide accuracies as high as
millimeters for applications such as surgical navigation [114-116]. For example, Kuhn et
al. [117] compared the performance of different algorithms for UWB based localization,
including RSSI, matched filter, iterative peak subtraction, and leading edge. The
simulation results showed that the leading edge algorithm yielded an average error of
3.30 mm in a 27 m
3
space, which was the best among all algorithms. However, such
research is outside the scope of this thesis due to its scale of environment and level of
accuracy, therefore a comprehensive review of such research is not included in this thesis.
5.2.4.4. WSN Based Solutions
A WSN is a group of spatially distributed autonomous sensors that monitor
environmental conditions. As an emerging RF technology, it has been actively used in
indoor localization, and a variety of algorithms has been proposed and tested [118,119].
For example, Zanca et al. [120] compared four localization algorithms that used RSSI
Chapter 5: Review and Evaluation of Indoor Localization Technologies
52

readings collected from WSN. These algorithms included Min-Max, multilateration,
maximum likelihood, and ROCRSSI. The maximum likelihood algorithm was reported to
be the optimal algorithm. Priwgharm and Cherntanomwong [121] did another
comparison, and concluded that range based lateration algorithm outperformed min-max
and fingerprinting algorithms. A range of other WSN based indoor localization solutions
have been reported in [122-130].
Calibration and integration of algorithms and technologies have been proposed to
improve performance of WSN based solutions. Dominguez-Duran et al. [131] proposed
to calibrate a signal propagation model used in range based localization by dynamically
estimating RSSI-distance relationship using RSSI readings between WSN nodes. Field
tests in three scenarios reported accuracies between 3 and 4 m. Chen and Juang [132]
introduced an outlier-detection technique to detect and eliminate abnormal RSSI readings,
which resulted in an improvement of accuracy by 13-30%. Mitilineos et al. [133]
proposed a WSN based solution that integrated three algorithms, namely plain RSSI,
scene-analysis RSSI and approximate point-in-triangulation, and an optimal fusion rule.
Field tests reported mean accuracies between 0.83 and 2.01 m for the three algorithms
and their fusion, with scene-analysis RSSI alone yielding the best accuracy. Xiong el al.
[134,135] used an extended Kalman filter to integrate RSSI data collected from WSN and
proximity data collected from RFID. Simulation showed that the integrated solution
outperformed individual solutions. Zhang et al. [136] proposed another hybrid solution
using RFID and WSN. In addition to the use of RSSI, they used the variance of RSSI in
location estimation, and reported an accuracy of 0.45 m in simulation.
Chapter 5: Review and Evaluation of Indoor Localization Technologies
53

5.3 Evaluation of Indoor Localization Technologies for Building
Emergency Response Operations
Since application areas of indoor localization solutions are different, design requirements
and evaluation criteria of these solutions also differ. For example, a room-level accuracy
is sufficient for a solution used in building lighting control, whereas a meter-level
accuracy is usually required for a solution used in locating equipment in facilities
management; or pre-collection of radio signal data is acceptable for a solution used in
serving visitors of museums, whereas it is undesired for a solution used in facilitating
building emergency response operations. Therefore, to evaluate indoor localization
technologies, it is important to specify the area of application, which is building
emergency response operation in this thesis. It is also important to base the evaluation on
requirements specifically applied to this application. Accordingly, all indoor localization
technologies reviewed in this thesis are evaluated against the requirements identified in
the survey and presented in chapter 4.2.2, in order to select the technology that is the
most competent for indoor localization solution. For a particular technology, while its
performance varies in different solutions, its best performance against a particular
requirement is used in the evaluation. The evaluation is summarized in Table 4.
INS has been proven capable of providing a room-level or higher accuracy, given short
travel distances in indoor environments. Its on-scene deployment includes equipping first
responders with inertial sensors, and capturing their initial locations to star tracking. It
has high resistance to heat, water, smoke and other physical damages, as inertial sensor
are attached to people and not fixed at locations that may be directly exposed to hazards
Chapter 5: Review and Evaluation of Indoor Localization Technologies
54

in environments. With recent development of IMUs, size and weight of INS devices can
well satisfy the maximum size and weight requirements. Computational speed of INS is
fast, as step detection and orientation recognition is relatively straightforward.
Judging from prior research, AGPS does not guarantee a room-level accuracy. It requires
special antennae to be installed on scene. These antennae are susceptible to heat and
water, leading to reduced robustness of AGPS. Commercial AGPS receivers are small in
size and light in weight that can meet the corresponding requirements. Computational
speed is fast after initial cold-starts, if signal reception is good in indoor environments.
Infrared technology can provide a room-level or higher accuracy. In order to locate all
first responders or trapped occupants, infrared sensors need to be installed in every room
and have line of slight to tracked persons, which may be difficult to satisfy on emergency
scenes. Infrared sensors need to be rugged and protected from heat, water, smoke and
other physical damagers. First responders and trapped occupants don’t have to wear any
devices. The computational load is usually small, and instant computation and
presentation of location information is possible.
RF technologies can provide a room-level or higher accuracy. A sensing network
comprised of a number of location-known RF nodes is required. For first responders and
trapped occupants to be located, they need to carry mobile RF nodes, which are usually
small in size and light in weight. Deployed fixed nodes must be protected from heat,
water and other physical damages, while mobile nodes get better protection from tracked
persons. Depending on localization algorithms, computational speed can be fast.
Chapter 5: Review and Evaluation of Indoor Localization Technologies
55

The evaluation indicates that AGPS and infrared are problematic in one way or another,
and therefore cannot satisfy all the indoor localization requirements for building
emergency response operations. INS and RF technologies have potential to meet these
requirements. A closer examination of the objectives of this thesis reveals, however, that
INS suffers from an inherent limitation: when a building emergency breaks out, even if
trapped occupants are assumed to have access to inertial sensors, it is impossible to
initialize an INS solution with the trapped occupant’s initial locations. In other words,
unlike RF based localization solutions, INS is designed to continuously track the location
of a person starting from his initial position. However, it is not capable of detecting the
initial position itself, and will lose track of that person once the tracking process is
interrupted for any reason. Due to such limitation, INS can be used in locating first
responders only, but not trapped occupants. RF technologies, therefore, have the most
potential to support indoor localization for building emergency response operation, and
they are used in this thesis.
Table 4: Best performance of indoor localization technologies against five requirements
Technology Requirements
Accuracy Ease of on-
scene
deployment
Resistance
to damages
Device
size and
weight
Computational
speed
INS Centimeter
level
No need for
external
infrastructure.
Carry-on
sensors well
protected
Can be
made
small
Algorithms
available that
can perform
Chapter 5: Review and Evaluation of Indoor Localization Technologies
56

Need to attach
sensors to
targets
and light calculation in
real time
AGPS Meter
level
Antennae are
difficult to
install and
calibrate.
Targets need to
hold receivers
Antennae
sensitive to
damages
Can be
made
small
and light
Cold-start may
take time to
locate satellites
and calculate a
GPS lock
Infrared Centimeter
level
Dense
deployment of
sensors is
needed. No need
to attach sensors
to targets
Resistant to
partial loss
of deployed
sensors
Can be
made
small
and light
Algorithms
available that
can perform
calculation in
real time
RF Millimeter
level
Ad-hoc sensing
network can be
established on
scene. Need to
attach sensors to
targets
Resistant to
partial loss
of deployed
sensors
Can be
made
small
and light
Algorithms
available that
can perform
calculation in
real time

5.4 Conclusions
This chapter provides a thorough review of existing indoor localizations solutions that are
based on various technologies including the INS, AGPS, infrared, and RF. The review is
followed by an evaluation and comparison of these indoor localization technologies
Chapter 5: Review and Evaluation of Indoor Localization Technologies
57

against the requirements identified in chapter 4. It is concluded that the RF technologies
are the most promising for developing indoor localization solutions to support building
emergency response operations. The RF technologies are therefore used in the indoor
localization framework that is explained in the following chapters.

Chapter 6: Design of Indoor Localization
Algorithms for Building Emergency Response
Operations
Compared to competing indoor localization technologies, RF technologies are found to
be the most promising in satisfying indoor localization requirements for building
emergency response operations. The indoor localization framework introduced in this
thesis is built on the RF technologies accordingly. Beginning with an overview of
different types of existing localization algorithms for the RF technologies, this chapter
introduces two novel localization algorithms, including an environment-aware radio
frequency beacon deployment algorithm for sequence based localization (EASBL)
algorithm and an iterative maximum likelihood estimation (IMLE) algorithm. These
algorithms are designed for two different situations at emergency scenes depending on
whether existing sensing infrastructure is available in the buildings. This chapter presents
the detailed design of these algorithms. It is followed by chapters 7 and 8 that present the
detailed evaluation of these algorithms.
6.1 Overview of RF Based Indoor Localization Algorithms
Algorithms that have been used with RF technologies for indoor localization can be
categorized as two types: range based algorithms and range free algorithms.
Range based algorithms use a radio signal propagation model to map RSSI values to
physical distances in location computation. Most widely used range based algorithms
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
59

include triangulation algorithms [75] and maximum likelihood estimation (MLE)
algorithms [137]. Triangulation algorithms are performed by lateration using distance
measurements between reference positions. Theoretical or empirical models are used to
map collected RSSI, TOA or TDOA to physical distances between transmitters and
receivers, and the distances are used for triangulation to obtain the 2D or 3D positions of
targets. MLE algorithms use a likelihood function that is based on location measurements
between reference nodes and a target node. Searching for a target location is performed
by maximizing the likelihood function.
There are a number of other algorithms that do not map RSSI values to a physical
distance between two nodes using a radio signal propagation model. These algorithms are
generally categorized as range free algorithms. Depending on techniques used in location
computation, range free algorithms include proximity algorithms, fingerprinting
algorithms, and sequence based localization (SBL) algorithms. Proximity algorithms
require measurements of nearness of a set of location-known neighboring points. When a
transmitter or a receiver is attached to a target, it continuously communicates with
receivers or transmitters deployed in the environment through radio waves. RSSI, TOA
or TDOA of the radio waves is observed and used to measure the relative nearness of
these reference positions. The measured nearness, along with corresponding known
locations, is used to estimate the location of the target. Fingerprinting algorithms capture
signal strength in a sensing area and compare it with a preexisting signal strength
database to map a target to its location. Pre-application mapping is required, during
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
60

which sensing infrastructure, such as RFID readers or tags, is deployed in a sensing area,
and signal strength of reference positions is measured and recorded to establish a radio
map. During the application, signal strength of radio waves that travel between a target
and the infrastructure is measured and compared with the radio map. Known locations of
reference positions that best match the database are then used to estimate location of the
target. The SBL algorithms divide a sensing space into a large number of small regions
with perpendicular bisectors of the lines joining every pair of nodes in a network. Each
distinct region can be uniquely identified with a location sequence that represents
distance ranks of all nodes to that region. When the location sequence of a target is
determined based on RSSI values of all nodes, the nearest location sequence in a location
sequence table is identified, and the centroid of the associated region is considered to be
the location of the target.
Despite their respective strengths, the above-mentioned algorithm types all bear certain
limitations in meeting all the needs by first responders. For example, range based
algorithms, which depend on a mapping between RSSI values and physical distances, can
be questionable, as the signal models are usually arbitrarily given and do not take into
account the impact of the environment on signal propagation. Among different range free
algorithms, fingerprinting algorithms are considered unsuitable, because of the need for a
large number of fingerprints. Collection and maintenance of these fingerprints are usually
time- and labor-intensive and are, therefore, not practical in most buildings. More
importantly, due to unpredictable changes on emergency scenes, such as irregular
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
61

movements of people or possible structural collapses, fingerprints may become invalid,
which in turn compromises the entire localization solution. Proximity algorithms also
have their drawbacks, as they usually require a high density of sensor grids to ensure a
high accuracy, and it may not be feasible to deploy and maintain sensor grids at chaotic
emergency scenes.
To satisfy the indoor localization requirements, while addressing the limitations of
existing algorithms, this thesis introduces two novel algorithms. If emergency response
operations are carried out in buildings, where there is no existing sensing infrastructure,
an ad-hoc sensor network needs to be established on scene. For such situations, THE
EASBL algorithm is introduced. Built on the SBL algorithm, the EASBL algorithm is
preferred over proximity algorithms due to one major advantage: it requiring less sensing
infrastructure while ensuring a high accuracy. This advantage reduces efforts required to
establish an ad-hoc sensor network on scene. Compared to existing SBL algorithms,
EASBL is featured by its awareness of the sensing environment with integration of
building information, and its capability of increasing room-level accuracy and reducing
deployment efforts with particular sensor deployment plans. Meanwhile, if there is a pre-
installed sensor network in buildings, as assumed by the majority of indoor localization
literature that uses RF technologies, the IMLE algorithm is introduced. The algorithm
makes use of the existing sensing infrastructure. Compared to other proximity algorithms,
the novelty of the IMLE lies in that it integrates a MLE method to make use of the
information presented by collected radio signal data, and introduces an iterative process
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
62

that takes into consideration impacts of obstructions (e.g. walls) on signal propagation.
The detailed design of these two algorithms is explained in the rest of this chapter.
6.2 The EASBL Algorithm
Sensing infrastructure needed for RF based indoor localization does not exist or is not
available to first responders in most of existing buildings. As a result, an ad-hoc sensor
network must be set up on scene. In such cases, the EASBL algorithm can be used for
localization.
6.2.1 Review of the Sequence Based Localization Schema
Sequence based localization is a range free indoor localization schema, first proposed by
Yedavalli and Krishnamachari in [123] and improved in [122]. At the heart of SBL is
division of a 2D space into distinct regions. Consider a 2D space that consists of n
reference nodes. For any two reference nodes, draw a perpendicular bisector to the line
joining them. This perpendicular bisector divides the localization space into three
different regions that are distinguished by their respective proximity to either reference
node. For n reference nodes, there are a total of ( 1) / 2 nn  pairs and hence ( 1) / 2 nn 
perpendicular bisectors, dividing the space into a number of regions. For each region, an
ordered sequence of the reference nodes’ ranks based on their distances to the region is
defined as a location sequence of that region. The location sequence is proven to be
unique for any given region [123], and is used as a unique identity of the region. The
space division process is illustrated in Figure 2.
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
63

Figure 2: Illustration of SBL (a) The perpendicular bisector of the line joining two
reference nodes divides the localization space into three distinct regions. (b) Illustration
of arrangement of six bisector lines for four reference nodes placed uniformly and
randomly in a square localization space. (c) Examples of location sequences for a four-
reference-node topology. (d) All feasible location sequences for the topology in (c). [123]
Once the space division is done, location sequences of all regions are recorded in a
location sequence table. Then, RSSI values of all reference nodes received by a target
node are used to form an ordered sequence, which is used as a location sequence of the
target node. The target node location sequence represents the ordered sequence of
distances between all reference nodes and the target node. Because of multipath and
fading effects, the target node location sequence usually is corrupted and it deviates from
the ideal location sequence. This is a major source of localization error.
Lastly, a location sequence in the location sequence table that is “nearest to” the target
node location sequence is identified, and centroid of the region represented by the
matching location sequence is used as an estimated location of the target node. The
nearness can be determined by Spearman’s rank order correlation coefficient [123].
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
64

Given two location sequences {}
i
Uu  and {}
i
Vv  , 1 in  , where
i
u and
i
v are the
ranks of reference nodes, Spearman’s rank order correlation coefficient  is defined as a
linear correlation coefficient of the ranks as provided by
2
1
2
6 ( )
1
( 1)
n
ii
i
uv
nn






(1)

The SBL schema is advantageous over or comparable to other localization schemas such
as proximity based localization, with its ability of providing a high coordinate-level
accuracy, requiring low number of reference nodes compared with other range-free
schemas, and being free of pre-data collection. Moreover, with n deployed nodes, it is
proven in [122] that there are ()
n
On possible location sequences, out of which only
4
() On are linked to regions and included in the location sequence table. The rest of the
location sequences, which account for the majority of all possible location sequences, are
considered invalid and not included in the location sequence table. When a target node
location sequence is corrupted due to multipath and fading effects and hence becomes
invalid, it is mapped to the nearest location sequence in the location sequence table. Due
to the low density of the valid location sequences, the nearest location sequence is likely
to be the correct location sequence. In this way, the SBL schema provides partial
mitigation to multipath and fading effects. Therefore, in this thesis, SBL is adopted for
location computation.
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
65

6.2.2 Algorithm Design
While the SBL schema has proven to have the above-mentioned advantages, its success
relies on the success of the space division, which is essentially determined by the
deployment of reference nodes. Generally, a higher accuracy can be expected when more
reference nodes are used and when they are more evenly distributed within a sensing area.
However, at emergency response scenes, where no sensor network is available, an ad-hoc
sensor network must be quickly set up. Therefore, the following challenges need to be
addressed: first, using a large number of reference nodes can ensure a high localization
accuracy, but at the cost of an increased deployment effort and hence a potential delay to
the emergency response operations; secondly, not all spaces may be easily accessed for
sensor deployment, and it is not always possible to evenly distribute sensor nodes to
maximize the accuracy. Some spaces that need to be covered by the ad-hoc sensor
network may not contain any reference nodes due to accessibility or other on-site
limitations.
The EASBL algorithm is introduced to address the above challenges. The algorithm is
designed to reduce the effort needed to cover a given sensing area by strategically
selecting locations for sensor deployment. It also integrates metaheuristics to search for
desirable sensor deployment that achieves a balance between the localization accuracy
and on-scene deployment effort. More importantly, the EASBL is BIM centered. The
close integration of BIM, which is a unique contribution of this thesis, is motivated by the
fact that the SBL schema provides coordinate-level location estimation, not room-level
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
66

location estimation. However, locations within the same region are not necessarily within
the same room. This may lead to false room-level estimations. In other words, even when
a coordinate-level accuracy is high, the corresponding room-level accuracy may be low.
The EASBL algorithm integrates BIM to increase the quality of the space division and
therefore aims to reduce the possibility of false room-level estimation. Moreover, the
knowledge of geometric information of an emergency scene, especially layout of walls
and doors, can help with decision-making about the layout of the ad-hoc sensor network.
The environment-awareness of the algorithm is demonstrated by its ability to recognize
and make use of spatial characteristics of an environment, and its utilization of building
information in sensor deployment optimization and target location computation [24].
6.2.2.1 Mathematical Formulation
In the EASBL algorithm, the quality of a space division is represented by a percentage of
the “breakaway area” referred to as
ba
p . In the SBL schema, the centroid of a region is
used as an estimated location of a target node within that region. However, part of the
region may be in a room different than the region centroid. This part of the region is
defined as a “breakaway area”. An incorrect room-level estimation occurs when a target
node is located within a breakaway area. Assuming the possibility of a target being at a
specific location is equal for any location within the sensing area, the percentage of such
incorrect estimations to all estimations is equal to the percentage of all breakaway areas
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
67

within the entire sensing area. Therefore, a smaller value of indicates higher quality
of space division and consequently a higher room-level accuracy.
On the other hand, the on-scene deployment effort is represented by the total number of
reference nodes n , and by the difficulty in deployment of each reference node. The
difficulty in deploying reference node i is measured by a penalty ,1
i
c i n  . At
emergency scenes, it is not possible to measure the exact coordinates of reference nodes
and communicate them between first responders and an incident commander. Therefore,
a set of m candidate locations for reference node deployments is chosen on the fly based
on the building layout. The location set includes two types of locations: (1) if reference
node i is to be deployed in a hallway close to a door, first responders can easily deploy it
when they pass by, and the penalty
i
c is set to be 1; (2) if reference node is to be
deployed at the center of an enclosed space, it usually requires first responders to break a
lock and get into the room. In this case,
i
c is set to 2 to penalize the increased effort. By
using these candidate locations that do not involve measurement of exact coordinates, an
incident commander can easily provide deployment commands to first responders, and
first responders can easily find the locations for reference nodes and carry out the
commands. Less reference nodes and less penalty for each node are desired, leading to a
reduced deployment effort and a faster deployment speed.
The solution for the optimal ad-hoc sensor network deployment is the one that minimizes
the total breakaway area and the penalty of all deployed nodes. From the computational
ba
p
i
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
68

point of view, this problem can be mathematically abstracted and expressed as: there are
m candidate locations chosen based on the building layout, and m reference nodes. Each
candidate location (1 ) i i m  can hold up to one node for a deployment penalty
i
c .
Each node can be deployed at either one of the candidate locations or none of them
(unused). For a given sensing area and given deployment of all nodes, a coverage penalty
ba
p can be calculated based on the sensor locations and building geometries, especially
walls and doors. The objective is to minimize the total penalty (TP):
   
,,
11
min
..
1, if location i is at hallway
{
2, if location i is at room center
1, if node jis deployed at location
{
0, othersize
1
1
01
0
ba i ij
mm
ba i ij
p c k
ij
i
ij
ij
i
ij
j
ba
TP e p c k
st
c
i
k
k
k
p
e











(2)

In the equation (2), e is a ‘tradeoff coefficient’ that represents the tradeoff between space
division quality and on-scene deployment effort, and
ij
k is a binary variable that denotes
whether a node j is deployed at candidate location i or not.
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
69

6.2.2.2 Integration of BIM and Metaheuristics
A contribution of the EASBL algorithm is its unique awareness about the environment
obtained by integrating building information, and its ability in achieving dual objectives
of accuracy and deployment efficiency supported by integrating metaheuristics in
searching the optimal deployment plan. The EASBL algorithm relies on BIM as a source
of building information. BIM is an approach to design, construction, and facilities
management, in which “a digital representation of a building that is used to facilitate the
exchange and interoperability of information in digital format” [138]. While most other
building information sources, such as drawings, images, or databases, may be more
accessible to first responders, these traditional building information sources contain only
geometric or semantic information, whereas BIM is a database of both. Although the
current accessibility of first responders to BIMs may be low, BIM deployment by the
architecture, engineering, construction and operation (AECO) industry as a central
repository of building information is increasing [139]. In the interviews with the LAFD
first responders, they also indicated that the availability of BIMs was increasing. This
thesis uses BIMs as the main building information source, and explores the utilization of
building information in indoor localization based on BIMs. Building information is used
in two essential and critical ways in the EASBL algorithm:
First of all, given a sensing area, building information available in BIM models,
especially the geometric information, is used to identify spaces, where people may be
located, including rooms, corridors and staircases. The scope of the spaces to be
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
70

identified is restricted by the localization sensing area specified by the incident
commander. The building information is as also used to evaluate the accessibility of
indoor spaces, based on the layout and usage of the spaces that can be retrieved from
BIM models, and the spread of the fire and smoke that can be marked and updated in the
BIM models according to reports from deployed first responders. Based on the room
accessibility information, the EASBL algorithm can automatically identify candidate
locations for the reference node deployment, determine the deployment penalty for each
candidate location, and prepare all input data that is needed for the computation of the
satisfactory deployment plan.
Second, building information lays the basis of calculating
ba
p for a given space division.
To compute
ba
p , the space is first divided by perpendicular bisectors only, forming a
primary space division. Then, building geometries are integrated, and they intersect with
the bisectors to create smaller regions, resulting in a secondary space division that has
more regions with smaller sizes. Every region in the secondary division is essentially
created by dividing a meta-region in the primary division. Every region in the secondary
division is traversed and compared to its meta-region, and it is recognized as a breakaway
area if its centroid is located in a different room than that of the centroid of the meta-
region. The percentage of these breakaway areas to the entire sensing area is
ba
p , which,
together with the associated deployment penalty, determines the quality of the space
division. The geometric information obtained from the BIM models is the basis of
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
71

evaluating the quality space divisions, and finding a satisfactory space division is the key
to achieving a high accuracy and low deployment effort.
Moreover, the integration of BIM also brings two added benefits which, although not
directly related to the function of the EASBL algorithm, are valuable to ensure the
success of the entire localization process. The first benefit is to use annotations in BIMs
such as room numbers to facilitate the communication between an incident commander
and deployed first responders, so that the latter can quickly follow commands to find a
specific location to deploy a node or rescue an occupant. The use of BIM also provides a
graphical user interface (GUI) for user interaction, through which incident commanders
can specify input such as inaccessible rooms and boundaries of a sensing area, and
monitor the updated locations of targets (i.e. deployed first responders and trapped
occupants) visualized in the context of a building layout.
In the EASBL, metaheuristics are integrated to decide on the optimal sensor deployment
plan. As a soft computing technique, metaheuristics are strategies that use readily
accessible, though loosely applicable, information to control problem solving in human
beings and machines [140]. A metaheuristic is an important part of the EASBL algorithm,
as computational speed is critical in time-sensitive emergency response operations. It
outperforms a brute-force search in that, when the number of candidate locations increase
linearly, the number of possible solutions increase at a factorial speed. The classic and
most widely used metaheuristics for optimization problems include genetic algorithm
(GA), simulated annealing (SA), and tabu search (TS) [141]. The GA locates optima
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
72

using processes similar to those in natural selection and genetics. The SA finds optima in
a way analogous to the reaching of minimum energy configurations in metal annealing.
The TS is a heuristic procedure that employs dynamically generated constraints or tabus
to guide the search for optimum solutions. It needs to be emphasized that none of these
metaheuristics are guaranteed to find the global optima. However, after a sufficiently
large number of evaluations, they usually yield satisfactory, though not necessarily global
optimal, solutions to a given problem. These solutions are hereafter referred to as
satisfactory solutions in this thesis.
6.3 The IMLE Algorithm
For situations where there is existing sensing infrastructure that first responders can
access in the building to collect RF data needed for localization, a second algorithm,
namely the IMLE algorithm, is introduced in this thesis. With the existing sensing
infrastructure, the algorithm no longer needs to be devised to reduce the deployment
effort of ad-hoc sensor networks, which leaves more room for the framework to satisfy
other indoor localization requirements such as accuracy and robustness. While the
availability of the needed sensing infrastructure remains a debatable assumption in
current building stock, this assumption, shared by the bulk of the indoor localization
literature, is becoming more realistic with the growing popularity of RF devices.
According to a recent report, the market of RF components for consumer electronics is
projected to triple its volume from 2012 to 2017 [142]. Such increasing prevalence of RF
devices provides the possibility of turning these devices into distributed sensor nodes that
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
73

can be used for collecting RF data needed for localization. It could be realized in the
future through development of the RF technologies, updates of building codes and fire
regulations and compliance of building owners to the codes/regulations.
6.3.1 Assumptions
When there exists sensing infrastructure in the building, it is assumed that the following
information is available to first responders, and can be used to lay the basis of location
estimation:
A set of RF transmitters { | 1,2,..., }
i
M M i m 

are deployed in the building, with known
locations {( , ) | 1,2,..., }
ii
x y i m  and transmit powers { | 1,2,..., }
i
Tx i m  . A set of
transceivers { | 1,2,..., }
i
N N i n  is also deployed as references, with known locations
{( , ) | 1,2,..., }
ii
u v i n  . More importantly, the BIM model of the building is available to
first responders. Based on the BIM models, a list of rooms within the sensing area can be
identified as {( , ) | 1,2,..., }
ii
r c i l  , where
i
c is the centroid of room
i
r , and l is the number
of rooms.
6.3.2 Algorithm Design
The iterative maximum likelihood estimation algorithm, as its name suggests, adopts an
iterative process for estimating targets’ locations based on a MLE method.
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
74

6.3.2.1 Estimation of Signal Model Parameter Values
Let { | 1,2,..., }
i
T t i k  be the set of targets. For an arbitrary target t , denote its
unknown coordinates as
00
( , ) xy . Its distance to transmitter
i
M , denoted as
i
d , can be
calculated based on the following classic signal propagation model [143]:
0
( ) 10 log( ) L d L d   (3)
where () Ld is path loss of signal strength [dB] in distance d [m],
0
L is reference signal
strength loss value [dB] for 1 m that could be surveyed, and  is path loss exponent.
When the impact of walls is taken into consideration, the model can be revised as:
0
( ) 10 log( ) * L d L d p w     (4)
where w is the number of walls, and p is the signal attenuation per wall.
The values of the parameters in the model are environment-specific. For building
emergency scenes, in particular, the parameter values are subject to the impact of the
emergency situations, and need to be estimated dynamically. Moreover, the immediate
environment of a transmitter may differ from that of other transmitters, resulting in
different signal model parameter values. As a result, the parameter value estimation needs
to be completed individually for each transmitter. The estimation is based on the data
collected from the transceivers and the known information of the sensing infrastructure,
and is based on a MLE method explained as follows.
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
75

For an arbitrary transmitter, its signal that reaches transceiver
i
N follows the revised
signal model:
0
( ) 10 log( ) *
i i i
L d L d p w      (5)
where 1,2,..., in  , and  is the variance of the signal strength following a Gaussian
distribution. In the n equations,
i
d and log( )
i
d can be calculated based on the coordinates
of the transmitter and the transceiver, ()
i
Ld can be calculated based on the signal strength
data reported by
i
N and the known transmit power, and
i
w can be obtained from the BIM
model given the locations of the transmitter and the transceiver. Based on the n equations,
define:
0 1 1 1
1 log( ) ( )
10
log( ) ( )
n n n
L d w L d
X y b
n d w L d p

    
    
  
    
    
    
，， (6)
The vector b , which represents parameter values, can be estimated as follows:
1
, ( )
TT
y Xb b X X X y

 (7)
6.3.2.2 Maximum Likelihood Estimation of Target Locations
For an arbitrary target t ,
2 2 2
00
( ) ( )
i i i
x x y y d     , where
i
d

is the distance from
00
( , ) t x y to ( , )
i i i
M x y . Reorganize the above equation as follows:
2 2 2 2 2
0
( ) 2
i i i i i i o i
x y d x y x x y y        (8)
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
76

For 1,2,..., 1 im  , minus the equation above with the equation
2 2 2 2 2
0
( ) 2
m m m m m m o m
x y d x y x x y y        (9)
and get:
2 2 2 2 2 2
00
2 ( ) 2 ( )
i i i m m m m i m i
x y d x y d x x x y y y           where 1,2,..., 1 im  .
Then, define the following matrices and vector based on these 1 m  equations:
2 2 2 2 2 2
1 1 1 1 1
0
0
2 2 2 2 2 2
11
1 1 1
2( ) 2( )
2( ) 2( )
m m m m m
m m m m
m m m m m m
x x y y x y d x y d
x
X y b
y
x x y y
x y d x y d

  
         
 

  
 



 
     


，， (10)
and vector b , which represents the location of the target t , can be estimated as follows:
1
, ( )
TT
y Xb b X X X y

 (11)
6.3.2.3 Iterative Process
As can be seen from the last equation, the estimation of
00
( , ) xy

requires
i
d , 1,2,..., im  ,
which can be calculated based on the signal model:
0
( ( ) * )/10
10
ii
L d L p w
i
d
 
 (12)
As explained above, the values of
0
L , p and  can be estimated with the MLE method.
However, as the room that contains the target is initially unknown,
i
w cannot be
calculated. An iterative process is therefore introduced to address this problem. The
process starts with an assumption that target t is in room
i
r (initially, 1 i  ). The numbers
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
77

of walls between target t and all transmitters are then counted in the BIM model, and
results are used for calculating
i
d . The location of the target is then estimated using the
MLE method. If the estimated location is within room
1
r , the estimation is a converged
estimation. The iteration process terminates, and reports the converged estimation as the
location of target t . Otherwise, the estimation is considered non-converged, and a fitness
value of this non-converged estimation is calculated, based on a certain fitness function.
In this case, the iterative process continues to assume the target is in the room that
contains the previous estimated location, and then repeats the above computation.
When the number of iterations exceeds a predefined limit, the above iteration starts over,
with an updated assumption that target t is in room
1 i
r

. Such iterative process continues,
until a converged estimation is found or when i exceeds l (the number of rooms). In the
former case, the converged estimation is reported as the estimated location of the target;
in the latter case, the target is estimated to be in the room that is associated with the
highest fitness value. A fitness function is used to evaluate the quality of the non-
converged solutions reported in the iterative process. It measures the deviation of the
solutions from possible convergence, namely from being in the room that is assumed to
contain the target. The above iterative process is illustrated in Figure 3. Based on the
design of the iterative process and the definition of converged estimation, this thesis
introduces three different fitness functions: (1) the negative of the distance from the
estimated location and to the room boundary; (2) the negative of the distance from the
estimated location to the room center; and (3) the negative of the distance from the
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
78

current estimated location to the previous estimated location reported in the last iteration.
The effectiveness of these fitness functions is evaluated in simulation in chapter 7.3.1.
11
11
define constant: =the maximum number of iterations
foreach ( )
{
1;
0;
;
;
do
{
()
{
;
0;
;
}
;
assume is in ;
apply signal model to estimate distances between and transmitters;
apply ML
i
i
ij
I
tT
i
j
rr
converge false
if j I
i
j
rr
j
tr
t










1
11
;
1
1
E to estimate location of , denoted as ;
get the room attribute of , denoted as ;
calculate fitness
()
{
;
}
} while (! & & )
if ( )
{
is in room . . ;
}
else
{
is i
ij
ij ij
ij
ij ij
ij ij ij
tt
tr
f
if r r
converge true
converge i l
converge
t r s t r r
t








[1, ]
[1, ]
n room s.t. max{ }
}
}
ij ij uv
ul
vI
r f f




Figure 3: Pseudo code for the iterative process
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
79

6.4 Conclusions
The ability to locate deployed first responders and trapped occupants accurately and
quickly is of significant importance to the success of building fire emergency response
operations. This chapter presents details of two novel algorithms, the EASBL and the
IMLE, that aim to provide first responders, especially incident commanders, with the
ability to set up an-hoc RF network and locate first responders and trapped occupants in
buildings. The EASBL algorithm is designed to retain the advantages of the SBL schema,
including the ability of providing a high coordinate-level accuracy, requiring low number
of reference nodes compared with other range-free algorithms, being free of pre-data
collection, and providing partial mitigation to multipath and fading effects. Furthermore,
the EASBL algorithm is designed to improve room-level localization accuracy and
reduce the deployment effort of an ad-hoc sensor network, by proposing metrics for space
division quality and deployment effort, and optimizing an objective function that
balances the tradeoffs between the above mentioned objectives. Metaheuristics are
integrated in the EASBL to efficiently search for a solution that, although not necessarily
global optima, provides a satisfactory solution within reduced computational time
compared with a complete search. The IMLE algorithm is designed to make use of
existing sensing infrastructure in buildings without requiring the setup of ad-hoc sensor
networks. The IMLE algorithm uses the MLE method to dynamically estimate the
propagation model of the signal sent by each individual transmitter, and to estimate the
location of the targets. The IMLE algorithm also introduces an iterative computational
Chapter 6: Design of Indoor Localization Algorithms for Building Emergency Response
Operations
80

process to improve the localization accuracy by mitigating the impact of signal
attenuations caused by walls. Building information models are integrated to extract
geometric information of the sensing area, support the computation of space division
quality, and provide a GUI for user interaction. These two algorithms are evaluated both
in simulation and field tests. The results are presented in chapters 7 and 8.

Chapter 7: Simulation Based Evaluation of the
Localization Framework
The indoor localization framework was evaluated in simulation, which allowed for
repeated implementation of localization processes and extensive evaluation of the
framework. This chapter provides the details of the simulation setup and two simulated
building fire emergency scenarios, and then presents the evaluated performance of the
framework based on simulation results.
7.1 Simulation Setup and Scenarios
A C# script was written to implement both the EASBL and the IMLE algorithms. The
script was compiled as a dynamic link library (DLL) file, and integrated into a
commercial BIM authoring tool as an add-on. The add-on takes user input, extracts
building geometries, performs space division optimization, and computes target locations.
Targets are entities with defined coordinates and room attributes randomly generated by
the script and distributed across a given sensing area. The processing of building
geometries was done using the built-in functions of the BIM tool. The add-on then
visualizes estimated locations on floor plans. A simulation tool was developed to
simulate different localization scenarios. The simulation tool generates a number of
targets in a sensing area and implements the two algorithms. The tool simulates the
following signal propagation model [144]:
0
1
( ) 10 log( ) ( )
P
p
L d L d WAF p 

   

,
Chapter 7: Simulation Based Evaluation of the Localization Framework
82

where () Ld is path loss of signal strength (dB) in distance d (m),
0
L is reference signal
strength loss in 1 m,  is path loss exponent, WAF is wall attenuation factor, p is
number of walls, and  is a Gaussian term for signal shadowing. The values of
0
L ,  and
WAF, and the standard deviation of  used in simulation, unless otherwise stated, were
55.0 dB, 4.7, 2.0 dB and 2.0 dB, respectively, and hereafter referred to as default
simulation settings. It needs to be noted that the simulation methodology bears an
inherent limitation; it relies on a number of assumptions, e.g. the magnitude of signal
attenuation per wall, that are not necessarily realistic. The simulation was used in this
thesis, in addition to the field tests explained in chapter 8, mainly because of the
controllability of environmental factors that would otherwise complicate the analysis, and
the possibility of performing extensive repetition that can lead to statistically reliable
results.
One representative floor of a typical office building on the University of Southern
California campus was used as a simulation test bed (Figure 4). It has an area of about
1800 m
2
, containing a total of 14 single occupancy offices, and 12 multi occupancy
conference rooms or labs. The hallway has a length of about 100 m comprised of four
linear segments. Two building fire scenarios of different scales were simulated. Both
scenarios were designed based on the suggestions from a number of first responders, and
were reviewed and verified by two incident commanders from the LAFD in terms of the
representativeness of the incidents and the scope of the sensing areas. In scenario 1
(Figure 4a), two single offices (red) were on fire. Occupants in these offices, all
neighboring offices, and offices and conference room that were across the hallway and
Chapter 7: Simulation Based Evaluation of the Localization Framework
83

had doors open to the hallway (orange) need to be evacuated. Due to the spreading
smoke, visibility in the hallway outside the offices (cyan) was low, resulting in an
increased risk to the first responders. The sensing area is color-coded in Figure 4a with a
size of 221 m
2
. In scenario 2 (Figure 4b), a fire started in a lab and soon spread to another
lab across the hallway (red). All labs on the east side of the floor were shut down for fire
attack and search & rescue (orange). Visibility in the hallway (cyan) was low due to
smoke. The sensing area is color-coded in Figure 4b with a size of 729 m
2
.
In either emergency scenario, two situations were simulated, including situation 1, where
no existing sensing infrastructure existed in the building, an ad-hoc sensing network was
required, and the EASBL algorithm was used; and situation 2, where existing sensing
infrastructure was available, and the IMLE algorithm was used. In situation 1, first
responders deployed an ad-hoc sensor network following the deployment plan given by
the EASBL algorithm. In situation 2, in order to be comparable with situation 1, the
number and location of pre-deployed transmitters used by the IMLE algorithm were the
same as in situation 1. A total of six transceivers were evenly deployed along the hallway.
The deployments of the devices in both scenarios and both situations are illustrated in
Figure 5.

Chapter 7: Simulation Based Evaluation of the Localization Framework
84

(a)

(b)
Figure 4: Simulation scenarios
Chapter 7: Simulation Based Evaluation of the Localization Framework
85

Figure 5: Illustration of device deployments in the field tests
7.2 Evaluation of the Framework with No Existing Sensing
Infrastructure
In a real-world scenario, the localization process starts with the definition of a sensing
area based on inputs by an incident commander. When the incident selects the EASBL
algorithm, the algorithm provides an optimal ad-hoc network deployment plan based on
Chapter 7: Simulation Based Evaluation of the Localization Framework
86

the defined sensing area and given building information. Once the first responders deploy
the network, the algorithm calculates and updates all target locations based on signal data
from the network. The data is collected by signal transceivers, such as smartphones,
carried by the targets. Figure 6 is an overview of the described localization process [24].

Figure 6: Flowchart of EASBL based indoor localization process
7.2.1 Evaluation of Metaheuristics
The metaheuristic plays a critical role in the EASBL algorithm, and the performance of
the metaheuristic, measured by the speed it converges to a solution and the quality of that
solution, is in turn impacted by the values assigned to its parameters. These parameters
are usually problem-specific; therefore, each metaheuristic is carefully tuned in this thesis
before it is evaluated. The GA has three parameters, namely population size, crossover
rate, and mutation rate; the SA has two parameters, namely initial temperature and
reduction ratio; the TS has one parameter, namely tabu list size. For each of these
Chapter 7: Simulation Based Evaluation of the Localization Framework
87

parameters, the values tested in this thesis and the selected values are summarized in
Table 5. For each tested value, the associated metaheuristic was run for 5,000 evaluations
for both of the simulated scenarios, and reported convergence speed and quality of
converged solution laid the basis of parameter value selection. It needs to be noted that
most of these parameters are continuous in their respective domains; however, it is only
feasible to test a number of discrete values that are distributed across domains. The
parameter values tested in this thesis were chosen based on the following guidelines: (1)
the parameter values are evenly distributed across the parameter domains (e.g. the
population size and the crossover rate of GA); (2) every value has a certain multiple
relationship with its predecessor (e.g. the mutation rate in GA, the initial temperature in
SA, and the tabu list size in TS); and (3) every value has a higher resolution than its
predecessor (e.g. the reduction ratio in SA). Random values between the tested values
were used for sensitivity analysis, and the results showed that the tested values were
generally representative. Therefore, the selected values were reasonably good, though not
necessarily the optimal in their respective domains.
Table 5: Summary of metaheuristics parameter values
Genetic
algorithm
Parameter Population size Crossover rate Mutation rate
Tested value set {50; 100; 150;
200; 250}
{0; 0.2; 0.4; 0.6;
0.8; 1}
{0.001; 0.005;
0.01; 0.05; 0.1;
0.5; 1}
Selected value 100 0.8 0.1
Chapter 7: Simulation Based Evaluation of the Localization Framework
88

Simulated
annealing
Parameter Initial temperature Reduction ratio
Tested value set {0.1; 1; 10; 100; 1000;
10000}
{0.9; 0.99; 0.999; 0.9999}
Selected value 100 0.999
Tabu
search
Parameter Tabu list size
Tested value set {1; 10; 50; 100; 500; 1000}
Selected value 10

Based on these selected parameter values, a comparison across the three metaheuristics
was carried out. For each scenario, all metaheuristics were tested for 10,000 evaluations,
and this process was repeated for 5 runs. The results showed that the TS performed
consistently better than the GA and SA, as it required fewer evaluations to converge to its
satisfactory solution, and the satisfactory solution had larger fitness values. Therefore, the
TS was integrated within the EASBL algorithm and used in the following analyses. In
scenario 1 and scenario 2, the TS specified that a total of 11 and 13 nodes needed to be
deployed, respectively. A sample run for both scenarios are shown in Figure 7.
Chapter 7: Simulation Based Evaluation of the Localization Framework
89

Figure 7: Convergence speed of three metaheuristics from a sample run in two scenarios
7.2.2 Evaluation of Localization Accuracy and Deployment Effort
The accuracy and deployment effort of the EASBL algorithm are examined in this
subchapter. For the RF beacon deployment, optimal deployment plans, computed by the
tabu search metaheuristic, are compared to random placements that involved the same
amount of beacons. For the location computation, estimated locations, provided by the
SBL schema, are compared to those provided by a classical proximity based schema,
which assumes the location of the nearest beacon to a target as the target’s location. The
two types of deployment plans and two sets of estimated locations were cross-combined
and evaluated. The combination with the optimal deployment plan and SBL schema
represents the EASBL algorithm. For each combination, a total of 100 targets were
generated and randomly distributed in the sensing area. The accuracy consisted of room-
level accuracy, measured by the ratio of correct room-level estimations to the total
number of targets, and coordinate-level accuracy, measured by the average error distance
of meter-level estimation of all targets. The deployment effort was measured by the total
penalty of all deployed nodes. This process was repeated 100 times, and the results are
Chapter 7: Simulation Based Evaluation of the Localization Framework
90

presented in Table 6. The accuracy and effort was computed each time, and their overall
confidence intervals were calculated at a 95% confidence level.
Table 6: Evaluation of localization performance
Performance metrics Combinations of beacon deployment plans and
localization schemas
Optimized placement Random placement
SBL schema
(EASBL)
Proximity
schema
SBL
schema
Proximity
schema
Scenario 1
Room-level accuracy (%) 87.0 ± 3.6 38.1 ± 4.6 80.7 ± 7.1 45.4 ±
10.5
Coordinate-level accuracy
(m)
1.78 ± 0.24 2.77 ± 0.13 1.73 ± 0.63 2.85 ±
0.49
Total deployment penalty 15.0 ± 0.0 15.0 ± 0.0 22.7 ± 2.4 22.3 ± 2.2
Scenario 2
Room-level accuracy (%) 87.2 ± 3.8 48.7 ± 5.2 80.4 ± 5.3 51.7 ± 7.9
Coordinate-level accuracy
(m)
1.57 ± 0.22 4.43 ± 0.28 2.03 ± 0.45 4.62 ±
0.31
Total deployment penalty 19.0 ± 0.0 19.0 ± 0.0 27.6 ± 2.5 27.4 ± 2.3

Chapter 7: Simulation Based Evaluation of the Localization Framework
91

It can be seen from the table that, compared to the random placement with SBL schema,
the EASBL algorithm (optimal placement with SBL schema) improved the room-level
accuracy by 6.3% in scenario 1 and 6.8% in scenario 2. In general, there was no statistical
difference between the amount of nodes deployed under the optimal placement and
random placement (with either location computation schema), however, the total
deployment effort with the EASBL algorithm was much lower, a reduction of 32.5% in
scenario 1 and 31.7% in scenario 2, compared with the random placement with SBL
schema. This indicates that the deployment plans prepared by the EASBL algorithm
relied more on easily accessible locations for beacon deployment. Coordinate-level
accuracy with optimized placement was not necessarily higher than that of the random
placement (with either localization computation schema), corresponding to the fact that it
is not considered in the objective function. Moreover, it needs to be pointed out that for
most of the targets that were incorrectly estimated at the room-level with the EASBL,
their estimated rooms were neighboring their actual rooms. This was the case, for
instance, for 84.5% and 82.7% of all incorrect room-level estimations with the EASBL in
scenario 1 and scenario 2, respectively. This finding suggests that even an incorrect
estimation could potentially be used to indicate in which multi-room zone a target may be
located [24].
The proximity schema performed significantly worse than the SBL algorithm in terms of
both room-level and coordinate-level accuracies, mainly due to two reasons: the
proximity schema cannot mitigate the multipath and fading effects in signal propagation,
while the SBL schema was proven to be able to provide partial mitigation [123]; and its
Chapter 7: Simulation Based Evaluation of the Localization Framework
92

success depends on a high density of devices which were not available in the simulated
scenarios.
The accuracy of the EASBL was further assessed against changes in the variance of
signal shadowing. In the default simulation settings, the standard deviation of  in the
signal model
0
1
( ) 10 log( ) ( )
P
p
L d L d WAF p 

   

was set at 2.0 dB. In order to evaluate
the performance of the EASBL under different variances of signal shadowing, which
could be observed in real-world environments, the value of the standard deviation was
varied from 0 dB to 6.0 dB, at a 0.5 dB increment. The respective accuracies were
simulated and calculated, and results are illustrated in Figure 8. The results showed that
the accuracy generally declined when the variance of the signal shadowing increased.
The accuracies remained above 80% when the variance of signal shadowing was below
3.0 dB and 3.5 dB in scenario 1 and scenario 2, respectively. The accuracy was less
sensitive to the changes of signal shadowing variance in scenario 2.

Figure 8: Accuracy of the EASBL under different variances of signal shadowing
Chapter 7: Simulation Based Evaluation of the Localization Framework
93

7.2.3 Evaluation of Robustness
An important requirement reported in the survey is the robustness of the localization
system. Ad-hoc sensor networks deployed at fire emergency scenes are subject to various
challenges such as fire, water, smoke and physical damages. Being able to retain the
localization function under these challenges is critical to the capability of the system to
fulfill first responders’ needs.
In this thesis, the robustness of the EASBL algorithm was examined by measuring the
changes in accuracy against partial loss of deployed nodes. In satisfactory solutions
presented in chapter 7.2.1, a total of 11 and 13 sensors needed to be deployed for scenario
1 and scenario 2, respectively. For each scenario, partial loss of deployed nodes was
simulated. Beginning with the optima, randomly selected nodes were taken out, one at a
time, and accuracy was reported based on the remaining deployed nodes. In order to
offset the impact of randomness in selecting the nodes to be taken out, for a specific
number of remaining nodes, the process was repeated 100 times. The minimum number
of deployed nodes to make sensing space division possible was 2 nodes. The results are
shown in Figure 9.
The results illustrated the robustness of the EASBL algorithm against partial loss of
deployed nodes. In the scenario 1, when 2 nodes, or 18.2% of all nodes, were lost, the
accuracy was barely impacted. The room level accuracy remained above 80% until the
loss increased to or exceed 5 nodes, or 45.5% of all nodes. Similarly in the scenario 2,
when 4 nodes, or 30.1% of all nodes, were lost, the accuracy was barely impacted. The
room level accuracy remained above 80% until the loss increased to or exceeded 7 nodes,
Chapter 7: Simulation Based Evaluation of the Localization Framework
94

or 53.8% of all nodes. Such robustness indicates some redundancy exists in space
division, suggesting that a high accuracy can still be possible with reduced quality of
space division. The redundancy also makes the EASBL algorithm a promising indoor
localization algorithm at building fire scenes, where partial loss of sensor nodes is highly
likely, if not inevitable [24].

Figure 9: Assessment of robustness against partial loss of deployed sensor nodes
7.2.4 Tradeoff between Onsite Deployment Effort and Localization Accuracy
A contribution of the EASBL algorithm is that it specifies a tradeoff between improving
the space division quality, hence localization accuracy, and reducing the effort required to
carry out the sensor deployment. The tradeoff between these two objectives can be
quantified and managed based on incident commanders’ discretion that reflects their
professional experience and judgment of on-scene situations on the fly. The tradeoff is
represented by the parameter in equation 1. This tradeoff is examined in this subchapter.
Chapter 7: Simulation Based Evaluation of the Localization Framework
95

When the e value is less than 10, the EASBL algorithm indicates that no reference nodes
should be deployed in order to maximize the fitness. When the e value is greater than 10
and as it increases, the relative importance of space division quality increases.
Consequently, the ( 1
ba
p  ) value increases, expectedly with an increase of total
deployment penalty. When the e value is sufficiently enough, further increasing the
value has little impact on the tradeoff. In this thesis, the satisfactory solutions for various
e values ranging between 10 and 4000 are calculated, and their associated space division
quality and deployment effort are plotted in Figure 10, fitted with a smoothing spline. As
can be seen in the figure, at small e values and hence small total deployment penalties,
increasing the deployment effort had little impact on the space division quality. As e
value increased, the space division quality started increasing, causing an increase in the
deployment penalty at the meantime, until it reached a point where adding more reference
nodes could no longer contribute to the space division quality. The results indicate that
trading additional deployment effort for higher space division quality or vice versa was
most efficient when the total deployment penalty varied between 10 and 20 in Figure 10.
It is important to note that e values associated with the deployment penalties within this
range are not the optimum e values. The selection of the e value is subject to the
judgment of the incident commander, who decides whether the importance of a higher
accuracy exceeds the importance of easy deployment or vice versa. The incident
commander can then select from a set of e values by reviewing their associated space
division qualities and deployment efforts.
Chapter 7: Simulation Based Evaluation of the Localization Framework
96

Figure 10: Tradeoff in scenario 1 and scenario 2
7.3 Evaluation of the Framework with Existing Sensing Infrastructure
In a real-world scenario, the localization process starts with the definition of a sensing
area based on the inputs by an incident commander. When the IMLE algorithm is
selected by the incident commander, geometric information of the building and
information about the existing sensing infrastructure, including the locations and
specifications (such as transmit power and signal frequency) of fixed transmitters and
transceivers, are retrieved from a BIM model. The algorithm first estimates the signal
models based on signal data collected by the transceivers. The algorithm then performs
the iterative computation to calculate and update locations of all targets based on signal
data collected at target locations. The data at target locations are collected by mobile
transceivers, such as smartphones, that are carried by the targets. Figure 11 is an
overview of the described localization process.
Chapter 7: Simulation Based Evaluation of the Localization Framework
97

Figure 11: Flow chart of IMLE based indoor localization process
In the simulation of both scenarios, the number and deployment plan of the transmitters
were determined to be the same as the number and deployment plan of the nodes in the
satisfactory solutions of the EASBL. In this way, the simulation setup was comparable
between the two algorithms. The number of the transceivers was six, twice the minimum
number of transceivers needed to implement the MLE method. The transceivers were
evenly distributed along the hallway within the sensing area.
7.3.1 Selection of Fitness Function
Based on the design of the IMLE algorithm, if no converged estimation is found in the
iterative process, a non-converged estimation that has the highest fitness value is used as
the estimated target location. In this case, the localization accuracy is highly impacted by
the design of the fitness function that is used to evaluate the quality of the non-converged
solutions. Based on the nature of the fitness value, i.e. a measurement of the deviations
from non-converged estimations to related possible convergence, this thesis introduces
Chapter 7: Simulation Based Evaluation of the Localization Framework
98

and evaluates the following three different fitness functions: (1) the negative of the
distance from the estimated location to the room boundary; (2) the negative of the
distance from the estimated location to the room center; and (3) the negative of the
distance from the current estimated location to the previous estimated location reported in
the last iteration. The evaluation of these fitness functions was done as described below.
The simulation tool generated a total of 100 targets that were randomly distributed in the
sensing area. The accuracy consisted of room-level accuracy, measured by the ratio of
correct room-level estimations to the total number of targets, and coordinate-level
accuracy, measured by the average error distance of meter-level estimation of all targets.
This process was repeated 100 times for each fitness function. The accuracy was
computed for each repetition of the simulation, and its overall confidence intervals were
calculated at a 95% confidence level. The results are presented in Table 7.
Table 7: Performance of three fitness functions
Fitness
function 1
Fitness
function 2
Fitness
function 3
Scenario 1
Room-level accuracy (%) 95.0 ± 2.5 92.0 ± 2.4 86.9 ± 4.2
Coordinate-level accuracy (m) 0.72 ± 0.10 0.72 ± 0.10 1.02 ± 0.23
Scenario 2
Room-level accuracy (%) 95.1 ± 2.7 87.9 ± 4.4 84.8 ± 6.2
Chapter 7: Simulation Based Evaluation of the Localization Framework
99

Coordinate-level accuracy (m) 0.84 ± 0.11 0.88 ± 0.12 1.21 ± 0.35

As can be seen in Table 7, fitness function 1 performed consistently better in both
scenarios than the other two fitness functions, by yielding statistically higher room-level
and coordinate-level accuracies. This result supports the intuition, as the projection of a
non-converged estimation on the room boundary is the nearest possible convergence to
the non-converged estimation. Fitness function 1 was therefore used in the following
simulations as well as the field tests.
Repeated simulations showed that the confidence interval of the possibility of a location
estimation to be converged was 78.2 ± 5.3% at a 95% confidence level. A further
examination was carried out comparing the accuracies of the converged and non-
converged estimations. The confidence intervals of the accuracies, calculated at a 95%
confidence level, are shown in Table 8. The results showed that the accuracy of non-
converged estimations, while expectedly lower than the accuracy of converged
estimations, was still promising, remaining over 77.1 ± 10.1%.
Table 8: Comparison of converged and non-converged estimations
Converged Non-converged
Scenario 1
Room-level accuracy (%) 98.1 ± 1.7 83.1 ± 9.5
Coordinate-level accuracy (m) 0.61 ± 0.11 1.19 ± 0.19
Chapter 7: Simulation Based Evaluation of the Localization Framework
100

Scenario 2
Room-level accuracy (%) 99.7 ± 0.6 77.1 ± 10.1
Coordinate-level accuracy (m) 0.74 ± 0.10 1.40 ± 0.21

7.3.2 Evaluation of Localization Accuracy
To better evaluate the performance of the IMLE, it is compared to two other algorithms
that were also implemented and evaluated in the simulation. The first algorithm is the
widely used MLE algorithm. It locates targets by MLE based triangulation, and does not
take into account the impact of building geometries on signal propagation. The second
algorithm integrates the widely used proximity based algorithm, assuming a target is
located at its nearest neighboring transmitter, and offsets the impact of signal attenuations
through walls with an iterative process. This algorithm is termed as the iterative
proximity (IP) algorithm in this thesis. These two algorithms were selected for the
comparison as both algorithms share important characteristics (MLE based computation,
an iterative process) with the IMLE algorithm, and at the meantime integrate existing
indoor localization algorithms (i.e., MLE and proximity) that have been extensively
investigated in the prior research. The simulation process for evaluating the accuracy of
the IMLE, MLE and IP algorithms was the same as the process used for evaluating the
fitness functions. The confidence intervals of the accuracies, calculated at a 95%
confidence level, are presented in Table 9.

Chapter 7: Simulation Based Evaluation of the Localization Framework
101

Table 9: Evaluation of the accuracy of the IMLE algorithm
IMLE MLE IP
Scenario 1
Room-level accuracy (%) 95.0 ± 2.5 64.8 ± 4.7 39.5 ± 4.5
Coordinate-level accuracy (m) 0.72 ± 0.10 1.26 ± 0.15 2.76 ± 0.14
Scenario 2
Room-level accuracy (%) 95.1 ± 2.7 82.5 ± 3.9 52.1 ± 5.7
Coordinate-level accuracy (m) 0.84 ± 0.11 1.28 ± 0.15 4.40 ± 0.28

In both scenarios, the IMLE undeniably outperformed the MLE and IP, with a room-level
accuracy of over 95% and a coordinate-level accuracy of over 0.84 m. The MLE yielded
a slightly lower coordinate-level accuracy. However, its room-level accuracy was
significantly lower, especially in scenario 1 (64.8 ± 4.7%). This indicated that the MLE
algorithm, lacking the ability to mitigate the impact of walls on signal propagation, was
not capable of ensuring a high room-level accuracy, especially in compact spaces, similar
to those in scenario 1, where a small error distance would be enough to lead to incorrect
room-level estimations. The estimations by the IP were the least accurate, with a room-
level accuracy of lower than 53% and a coordinate-level accuracy of lower than 2.76 m.
This was probably due to the fact that the success of the proximity based localization
heavily relies on a high density of deployed transmitters, which not only would be over-
Chapter 7: Simulation Based Evaluation of the Localization Framework
102

demanding in most real-world emergency scenarios, and also would be likely to limit the
robustness of the algorithm against possible damage and loss of devices.
Moreover, it needs to be noted that the performance of the IMLE was generally
consistent between the two scenarios, with a variation of 0.1% for the room-level
accuracy and 0.12 m for the coordinate-level accuracy. The performance of the MLE and
the IP, to the contrary, was inconsistent. The variation of the room-level accuracy
between the two scenarios reached 17.7% and 12.6% for the MLE and IP, respectively,
and the variation of the coordinate-level accuracy reached 1.64 m for the IP. It is also
important to note that for 86.9% and 87.3% of the targets that were incorrectly estimated
at the room-level with the IMLE, their estimated rooms were neighboring their actual
rooms.
Similar to the EASBL, the accuracy of the IMLE was further assessed against changes in
the variance of signal shadowing. In the simulation, the standard deviation of the signal
shadowing was varied from 0 dB to 6.0 dB, at a 0.5 dB increment. The respective
accuracies were simulated and calculated, and results are illustrated in Figure 12. The
results showed that the accuracy generally declined when the variance of the signal
shadowing increased. The accuracies remained above 80% when the variance of signal
shadowing was below 4.0 dB and 4.5 dB in scenario 1 and scenario 2, respectively. The
patterns of the sensitivity of accuracy to the changes of signal shadowing variance were
similar in both scenarios.
Chapter 7: Simulation Based Evaluation of the Localization Framework
103

Figure 12: Accuracy of the IMLE under different variances of signal shadowing
7.3.3 Evaluation of Robustness against Loss of Transmitters
Damages to devices caused by hazards at emergency scenes are a main threat to an
indoor localization solution. Accordingly, robustness was identified as one of the major
requirements by first responders surveyed across the country. To evaluate the robustness
of the IMLE, the impact of partial loss of the transmitters, which are responsible for
generating RF signals that lay the basis of location computation, is analyzed in this
subchapter. In the simulation, beginning with 11 and 13 installed transmitters in scenario
1 and scenario 2, respectively, randomly selected transmitters were taken out, one at a
time, until only 3 transmitters were left, which was the minimum number of transmitters
needed to implement the MLE method. After the removal of each transmitter, the
accuracy was reported based on the remaining transmitters. In order to offset the impact
of randomness in selecting the transceivers to be taken out, for a specific number of
remaining transmitters, the process was repeated 100 times. The results are shown in
Figure 13.
Chapter 7: Simulation Based Evaluation of the Localization Framework
104

Figure 13: Assessment of robustness against partial loss of existing transmitters
The results showed promising robustness of the IMLE against partial loss of the
transmitters. In scenario 1, when 4 transmitters, or 36.4% of all transmitters, were lost,
the accuracy was barely impacted, with the room-level accuracy remaining over 90% and
the coordinate-level accuracy remaining over 1.0 m. Similarly, in scenario 2, when 6
transmitters, or 46.2% of all transmitters, were lost, the accuracy was barely impacted,
with the room-level accuracy remaining over 95% and the coordinate-level accuracy
remaining over 1.0 m. In both scenarios, the algorithm yielded acceptable accuracies
(room-level accuracy over 80% and coordinate-level accuracy over 2.0 m) with as few as
4 transmitters. The results indicated that the IMLE algorithm is capable of providing a
high localization accuracy with a low density of existing infrastructure, leading to not
only improved robustness but also extended applicability of the algorithm in practice.
7.3.4 Evaluation of Robustness against Loss of Transceivers
The fixed transceivers, which are part of the existing sensing infrastructure and used for
estimating the parameter values of the signal model, are also subject to damages and
Chapter 7: Simulation Based Evaluation of the Localization Framework
105

losses due to challenging environments at building emergency scenes. This thesis also
examines the robustness of the IMLE algorithm against the loss of the transceivers. For
each of the two scenarios, partial loss of the deployed transceivers was simulated.
Beginning with 6 installed transceivers, the process of repeatedly simulating the removal
of transceivers and assessing the resulting accuracy was the same as the process used in
evaluating the robustness against partial loss of the transmitters. Because there were 3
parameters in the signal model that needed to be estimated, a minimum of 3 transceivers
were needed to make the MLE based parameter value estimation possible. The results are
shown in Figure 14.

Figure 14: Assessment of robustness against partial loss of existing transceivers
In scenario 1, the room-level accuracy was reduced by 1.9% and the error distance was
increased by 0.13 m, when two out of the six transceivers were removed in the simulation.
In scenario 2, with the same loss of transceivers the room-level accuracy was reduced by
2.3% and the error distance was increased by 0.22 m. The results indicated measurable
robustness of the IMLE algorithm against partial loss of the transceivers. The impact of
Chapter 7: Simulation Based Evaluation of the Localization Framework
106

the loss of transceivers on the accuracy became significant when a third transceiver was
removed. With only three remaining existing transceivers, the room-level accuracy was
reduced to around 85% in both scenarios, and the error distance was increased to up to
2.04 m. The results suggested reduced reliability of the signal model parameter values
estimated using the minimum required number of transceivers.
7.4 Conclusions
The EASBL and IMLE algorithms are evaluated in two simulated building fire scenarios
in this chapter. For the EASBL, the simulation results showed respective room-level
accuracies of 87.0 ± 3.6% and 87.2 ± 3.8%, and respective error distances of 1.78 ± 0.24
m and 1.57 ± 0.22 m, all at 95% confidence level, for scenario 1 and scenario 2. This
thesis also evaluates the robustness of the localization algorithm. The overall deployment
effort was reduced by 32.1% compared with random placement of RF sensors in
establishing ad-hoc networks, indicating the effectiveness of the EASBL algorithm is
achieving the two intended objectives, namely improving accuracy and reducing
deployment effort. For the IMLE, the simulation results showed respective room-level
accuracies of 95.0 ± 2.5% and 95.1 ± 2.7%, and respective error distances of 0.72 ± 0.10
m and 0.84 ± 0.11m, all at 95% confidence level, for scenario 1 and scenario 2. The
results are promising, demonstrating the potential of the two algorithms in providing the
much needed access to location information at building emergency scenes.

Chapter 8: Field Test Based Evaluation of the
Localization Framework
Simulation based evaluation is advantageous in that the evaluation is highly repeatable,
and that the evaluation is carried out in a controlled environment so that the impact of a
particular factor can be isolated and analyzed. However, simulation based evaluation is
also challenged by the fact that the real-world environments are more complex and
unpredictable, which may impact the performance of the indoor localization framework
in ways that are unobservable in the simulation. To provide a more comprehensive
evaluation, the framework was evaluated in field tests. The design, implementation and
findings of the field tests are reported in this chapter.
8.1 Prototype Development
A prototype was developed for the field tests. The prototype consisted of three major
components: RF transmitters, smartphones, and a localization server.
The RF transmitters were off-the-shelf programmable routers (Figure 15a). Each
transmitter, with a size of 113 mm x 138 mm x 29 mm and a weight of 230 grams,
contained a 400MHz processor, 32MB onboard memory, two built-in high power
antennae, five Ethernet ports, and light-emitting diodes (LED) indicator lights. With a
transmit power of 30 dBm, the transmitter could create a 802.11b/g/n wireless access
point that was detectable up to fifty meters away in an open space. During the field test,
in scenarios where an ad-hoc network was needed, the transmitters were used as beacon
Chapter 8: Field Test Based Evaluation of the Localization Framework
108

nodes in the network, and they were deployed by first responders at the beginning of the
field test following the deployment plans developed by the EASBL algorithm. In
scenarios, where existing sensing infrastructure was available, the transmitters were
preinstalled in the test bed before the field test began, and their information, including
their mac addresses, service set identifiers (SSID), and locations, was recorded
beforehand and made available for location computation. In either case, each transmitter
had its own unique mac address and SSID, which were visible to any device that received
the RF signal the transmitter broadcasted.

(a) (b)
Figure 15: Transmitter (a) and smartphone (b) used in the prototype
The smartphones used in the prototype were also off-the-shelf smartphones (Figure 15b)
with a size of 116 mm x 60 mm x 14 mm and a weight of 169 g. Each smartphone
contained a 600 MHz processor, 256 MB memory, extendable storage capacity, and
built-in cellular, wifi and Bluetooth modules, and relied on a built-in battery for power. It
Chapter 8: Field Test Based Evaluation of the Localization Framework
109

supported Internet browsing, as well as customized and third-party applications. A
localization application was developed for this prototype (Figure 16) and installed in the
smartphones. The application, when run by the user, would turn on the wifi sensor in the
smartphone, scan all detectable wifi access points in the environment every five seconds,
collect RF signal data, and forward it to a localization server, whose IP address was
programmed into the application beforehand but could also be overwritten by user input.
The date sent by the application to the server included the mac address, SSID and RSSI
of every detected wifi access point, the device ID of the phone, and a timestamp. The
data transition could rely on either the wifi or the cellular connection. The application
also had a user interface that allowed users to monitor the data collection and transition
processes, and to stop or resume the localization function whenever needed. In addition,
an off-the-shelf time synchronization application was installed in the smartphone. It kept
the clocks synchronized among all smartphones and the localization server during the
field test. A few smartphones were pre-deployed and used as transceivers in the field tests.

Figure 16: Interface of the localization application
Chapter 8: Field Test Based Evaluation of the Localization Framework
110

A remote localization server was also set up for the prototype. Accessible to the incident
commander, the localization server consisted of four major components: a webserver, an
SQL database, a BIM platform, and a location computation module. The webserver was
responsible for receiving the signal data that were sent by the smartphone application in a
JSON format, parsing the data, and passing the data to the SQL database. The SQL
database, upon receipt of the data, checked the integrity of the data, and stored it in two
separate tables, one for real-time processing, and the other for data backup. The SQL
database was also used for maintaining additional data required for location computation.
This data included the location, mac addresses, SSID and transmit power of existing
transmitters and transceivers, the mapping between IDs of the phone users and the phone
device IDs. The SQL database also included a table that kept a record of all location
computation results. This table not only allowed further analysis of the localization
results, but also had the potential to be used for supporting an extended target tracking
function. The BIM platform, based on a commercial BIM authoring tool and a
customized add-on development, was used to interact with the BIM model of the test bed
building that had emergency situations. The BIM platform extracted building geometric
information from BIM models, and used it for interpreting the layout of existing sensing
infrastructure or the ad-hoc networks as well as supporting location computation. It also
provided a GUI that allowed users to define the sensing area, indicate the availability of
existing sensing infrastructure, monitor location computation progress, and see visualized
localization results. Lastly, the location computation module was responsible for
processing the data retrieved from the SQL database and the BIM models, estimating
targets’ locations by implementing the EASBL and IMLE algorithms, and sending the
Chapter 8: Field Test Based Evaluation of the Localization Framework
111

location computation results back to the SQL database for record and to the BIM
platform for visualization. Figure 17 shows the connection of the above components of
the prototype and the data flow in the field tests.

Figure 17: Data flow in the field tests
8.2 Field Test Scenarios, Procedures and Test Bed Setup
The field tests were carried out in the same building that was used as the test bed in the
simulation. Two imaginary building fire emergency scenarios were used in the field tests.
The size and layout of the test bed, as well as the scope of the fire and the boundary of
the localization sensing area are explained in chapter 7.1. Based on the resource dispatch
rule introduced in chapter 3, the test bed building falls within category B, and the
following resources would be dispatched upon the receipt of an emergency call: four
engine companies, two truck companies, two rescue ambulances, one battalion chief, and
one emergency medical service captain. In both scenarios, based on the discussions with
incident commanders from the LAFD, the following assumptions about the resource
allocation and the actions taken by first responders and trapped occupants within the
Chapter 8: Field Test Based Evaluation of the Localization Framework
112

sensing area were made: the truck companies, ambulances, battalion chief and medial
captain would be working from outside the building. Three engine companies would be
deployed to work inside the building, with a fourth engine company standing by. Each
engine company consisted of four first responders. Two deployed engine companies
would be assigned to attack the fire, with three first responders in each company working
together within the sensing area, and one first responder working outside the area on
water supplies. The teams would first lay out hoses, and deploy an ad-hoc sensor network
if needed. They would then work in or around burning rooms until the fire was put out.
One deployed engine company would be assigned to search and rescue trapped occupants.
The team would traverse all rooms in the sensing are and, when trapped occupants were
found, escort them to a safe zone. In both scenarios, it was assumed that five occupants
were trapped in four rooms when the first response teams arrived. Two of them would
move around, trying to find their way out but blocked by fire and smoke, until they were
found by the first responders. Three of them would stay in their rooms and wait for the
rescue. The emergency situation was assumed to be under control within 20 minutes after
the first response teams were deployed. To mimic this process, each field test was
designed to take a maximum of 20 minutes.
Similar to the simulation, in either emergency scenario, two situations were tested,
including situation 1, where no existing sensing infrastructure existed in the building, an
ad-hoc sensing network was required, and the EASBL algorithm was used; and situation
2, where existing sensing infrastructure was available, and the IMLE algorithm was used.
The layouts of the devices in both scenarios and both situations, the same to those used in
Chapter 8: Field Test Based Evaluation of the Localization Framework
113

the simulation, are illustrated in Figure 5. For each combination of scenario and situation,
a field test was designed and repeated twice. A total of eight field tests were conducted.
A total of sixteen test subjects participated in the field tests. One subject assumed the role
of a battalion chief, who operated the localization solution and commanded the
emergency response operation from outside the sensing area. Among the rest of the
fifteen subjects, ten assumed the role of first responders in three engine companies, and
five assumed the role of trapped occupants. These fifteen subjects acted as targets in the
field tests, and they were always monitored and located when they were inside the
sensing area during the field tests. Every target was equipped with a smartphone, which
installed the localization application and was synchronized with the localization server,
and a stopwatch, which was started when a field test started. Every target was given a
unique script that listed for every field test a number of locations he/she needed to
traverse. A target was instructed to take an action every 15 seconds paced by the
stopwatch. An action can be either moving to the next location indicated in the script, or
remaining at the current location. A target needed to perform a total of 60 actions in
every field test, including actions taken by a group of first responders in situation 1 to
deploy ad-hoc networks. There were 66 different locations that a target might be
instructed to visit. Scattered in the sensing area, these locations were marked with
numbered sticky notes for easy recognition, and their locations were measured
beforehand and used as the ground truth. All subjects went through one-hour long
training before the field tests. During the training, they were explained the purpose and
procedure of the field test, their respective roles, and the instructions in their individual
Chapter 8: Field Test Based Evaluation of the Localization Framework
114

scripts. They were also instructed about how to use the two applications installed in their
smartphones. In addition, the subjects were walked through the test bed during the
training to get familiar with the locations of the sticky notes, so that they could quickly
navigate in the building during the field tests.
8.3 Evaluation of the Framework with No Existing Sensing
Infrastructure
8.3.1 Localization Accuracy and Deployment Effort
The tabu search, which outperformed other metaheuristics in the simulation, was used in
the EASBL algorithm. An optimal tabu size of 10 as reported in the simulation was used.
For each scenario, the average accuracies of the two repeated field tests are presented in
Table 10. As the field tests could not be repeated as many times as the simulation, the test
results presented in Table 10 are averages of the results from both repetitions instead of
confidence intervals. The localization accuracy with a proximity schema, using the same
node placement, was also calculated, and the results are presented in Table 10 for
comparison.
In all of the four situation 1 field tests, the room level accuracy was above 80% and the
coordinate-level accuracy was above 2.5 m for the EASBL. The respective average room-
level accuracies in scenario 1 and scenario 2 were 82.8% and 83.7%, and the respective
average coordinate-level accuracies in scenario 1 and scenario 2 were 2.12 m and 2.29 m.
These results were significantly better than those with the proximity schema, suggesting
the competence of the EASBL. Moreover, further analysis revealed that 41.7% of
Chapter 8: Field Test Based Evaluation of the Localization Framework
115

incorrect room-level estimations with the EASBL fell within rooms neighboring the
correct rooms. This finding suggests that even an incorrect estimation could potentially
be used to indicate in which multi-room zone a target may be located. This is important,
especially when first responders fail to find trapped occupants in estimated rooms in the
first attempt and need clues for secondary places to search. In general, the accuracies
reported in the field test provided convincing evidence about the capability of the EASBL
in providing reliable location information at emergency scenes.
Table 10: Localization accuracy of the EASBL in the field tests
Performance Metrics EASBL Optimized placement
and proximity schema
Scenario 1
Room-level accuracy (%) 82.8 31.7
Coordinate-level accuracy (m) 2.12 3.23
Scenario 2
Room-level accuracy (%) 83.7 39.3
Coordinate-level accuracy (m) 2.29 3.06

Chapter 8: Field Test Based Evaluation of the Localization Framework
116

8.3.2 Robustness against Partial Loss of Deployed Nodes
Further analysis of the test data was carried out to evaluate the robustness of the EASBL
against occurrence of loss of deployed nodes. The ad-hoc networks deployed in the test
bed included 11 and 13 sensors in scenario 1 and scenario 2, respectively. For each
scenario, beginning with the initial deployment, randomly selected nodes were taken out,
one at a time, and the accuracy was re-calculated based only on the data collected from
the remaining nodes. In order to offset the impact of randomness in selecting the nodes to
be taken out, for a specific number of remaining nodes, the process was repeated 100
times. The minimum number of deployed nodes to make sensing space division possible
was 2 nodes. The results are shown in Figure 18.

Figure 18: Assessment of robustness against partial loss of deployed sensor nodes
In scenario 1, when no more than 4 nodes, or 36.4% of all nodes, were removed, the
room-level accuracy remained above 70% and the coordinate-level accuracy remained
above 2.5 m. In scenario 2, when no more than 7 nodes, or 46.2% of all nodes, were
removed, the room-level accuracy remained above 70% and the coordinate-level
Chapter 8: Field Test Based Evaluation of the Localization Framework
117

accuracy remained above 3.0 m. Such limited impact of device losses on the localization
accuracy provided a demonstration of the robustness of the EASBL algorithm in the field
test. However, when the devices losses continued to increase, the impact became more
obvious. A significant decline of accuracy was observed when 6 or more nodes were
removed in scenario 1, resulting in a room-level accuracy of below 60% and a
coordinate-level accuracy of below 3.0 m, and when 8 or more nodes were removed in
scenario 1, resulting in a room-level accuracy of below 60% and a coordinate-level
accuracy of below 3.5 m.
8.3.3 Ease of On-Scene Deployment, Computational Speed, and the Size and Weight
of Devices
In addition to the accuracy and robustness, the survey results reported in chapter 4 also
revealed that a localization solution should have ease of on-scene deployment, have fast
computational speed, and use devices with desirable size and weight. The ease of on-
scene deployment, measured by the deployment effort of ad-hoc networks, is improved
by the inherent design of the EASBL algorithm, and the effectiveness has been
demonstrated by the comparison of total deployment effort between deployment plans
generated by the EASBL, in both the simulation and the field tests, and those by random
placement, as discussed in chapter 7.2.2. During the field tests, the ten subjects assuming
the role of first responders were able to identify the specified locations for node
deployment and set up an entire ad-hoc network within 90 seconds after the instructions
were provided. While admittedly more time would be required undoubtedly for the ad-
hoc network deployment at real-world emergency scenes due to conditions such as fire
Chapter 8: Field Test Based Evaluation of the Localization Framework
118

and smoke, the results from the test deployment were promising. Well trained first
responders are likely to be able to complete the deployment within 135 s, which is the
maximum amount of time allowed to be spent on the deployment, as reported in the
survey in chapter 4.2.
In terms of the computational speed, it was observed that updating the location
estimations once, which involved collecting the data from the ad-hoc network, computing
the targets’ locations, and presenting the localization results in the BIM platform, took
less than 5 s, far less than 40.34 s, which was reported as an appropriate amount of
computational time in the survey. Moreover, the tabu search was used by the EASBL
algorithm to quickly search for satisfactory solutions to initial ad-hoc networks
deployment. The integration of this soft computing technique significantly improved the
computational efficiency. However, the EASBL algorithm, in its current implementation,
required excessive computational time in computation of the node deployment plan due
to the delay in processing of building geometries, which was performed by a BIM
authoring tool. The BIM tool was selected solely based on its availability regardless of its
efficiency, as the scope of this thesis does not focus on improving the computational
efficiency of processing building geometries. It was observed that the building geometry
processing was responsible for over 90% of the total computational time. Since the
processing of geometries occurs only when finding satisfactory solutions to node
deployment, the impact of low efficiency in geometry process was limited to the
impediment of fast network deployment at the beginning of the emergency response
operation.
Chapter 8: Field Test Based Evaluation of the Localization Framework
119

Lastly, the sizes of the transmitters (0.45 cm
3
) and the smartphones/transceivers (0.10
cm
3
) were below the 107.34 cm
3
threshold identified in the survey. The weights of the
transmitters (0.23 kg) and the smartphones/transceivers (0.17 cm
3
) were also below the
1.16 kg threshold identified in the survey.
8.4 Evaluation of the Framework with Existing Sensing Infrastructure
8.4.1 Localization Accuracy
In two field tests conducted for scenario 1, situation 2, the IMLE algorithm achieved an
average accuracy of 84.6% at the room level and 1.85 m at the coordinate level. In two
field tests conducted for scenario 2, situation 2, the IMLE algorithm achieved an average
accuracy of 86.2% at the room level and 2.07 m at the coordinate level. The results are
summarized in Table 11. The accuracies were consistent across the two scenarios that
involved sensing areas of different sizes and shapes. Test results also showed that 62.5%
of incorrect room-level estimations fell within neighboring rooms, suggesting that a
secondary estimation of a target’s locations could be possible if the initial estimation
turned out to be incorrect.
Table 11: Evaluation of the accuracy of the IMLE algorithm in the field tests
Performance Metrics Localization Algorithm
IMLE MLE IP
Scenario 1
Room-level accuracy (%) 84.6 51.7 27.8
Chapter 8: Field Test Based Evaluation of the Localization Framework
120

Coordinate-level accuracy (m) 1.85 3.98 7.49
Scenario 2
Room-level accuracy (%) 86.2 55.3 32.5
Coordinate-level accuracy (m) 2.07 4.06 7.17

Furthermore, the MLE and IP algorithms were implemented based on the signal data
collected from the field tests. The design of the MLE and IP are explained in chapter
7.3.2. The results, which are also summarized in Table 11, showed that the accuracies of
the MLE and IP were significantly lower than those of the IMLE at both the room level
and the coordinate level in both scenarios. The failure of the MLE was due to its
negligence of RF signal attenuations caused by walls, and the failure of the IP was due to
its heavy reliance on a high density of transmitters that was unavailable in the field tests.
This comparison highlights the advantages of the IMLE, particularly its capability of
offsetting the impact of walls and its insensitivity to the amount of devices.
8.4.2 Robustness against Partial Loss of Existing Transmitters
The test data were further analyzed to evaluate the robustness of the IMLE against
occurrence of losses of devices, including transmitters and transceivers. Beginning with
the initial deployment, the analysis process was the same as the process for analyzing the
robustness of the EASBL. A minimum of 3 transmitters and 3 transceivers were needed
for implementing the IMLE algorithm. This subchapter discussed the findings about the
Chapter 8: Field Test Based Evaluation of the Localization Framework
121

robustness against loss of transmitters. The results are shown in Figure 19. The findings
about the robustness against loss of transceivers are discussed in subchapter 8.4.3.

Figure 19: Robustness against partial loss of existing transmitters in the field test
As can be seen in Figure 19, the IMLE was proven robust against partial loss of
transmitters. In scenario 1, the room-level accuracy remained above 80% when up to 3
transmitters, or 27.2% of all transmitters, were removed. Similarly, in scenario 2, the
room-level accuracy remained above 80% when up to 4 transmitters, or 30.8% of all
transmitters, were removed. The impact of loss of transmitters became dominant when
only 4 or fewer transmitters were left. The results suggested that a satisfactory accuracy
could be achieved with a small number of transmitters, and that extra transmitters added
to the network, while not making significant contribution to the accuracy, could provide
additional robustness to the localization solution.
Chapter 8: Field Test Based Evaluation of the Localization Framework
122

8.4.3 Robustness against Partial Loss of Existing Transceivers
Similar to the simulation, it was found in the field test that in both scenarios the room-
level accuracy was not significantly impacted by the loss of 2 or fewer transceivers, with
the accuracy remaining above 75% and 80% in scenario 1 and scenario 2, respectively.
However, unlike the simulation, the field tests reported large reductions in the accuracy
to an unreliable level of below 55% when a third transceiver was removed. The impact
of the loss of transceivers was highly consistent across the two scenarios.

Figure 20: Robustness against partial loss of existing transceivers in the field test
8.4.4 Ease of On-Scene Deployment, Computational Speed, and the Size and Weight
of Devices
In addition the provision of satisfactory accuracy and robustness, the IMLE was proven
to be capable of satisfying the other three important requirements identified in the survey.
First of all, the IMLE, unlike the EASBL, does not require computation of ad-hoc
network deployment plans and the deployment of the networks. Therefore, it could be
Chapter 8: Field Test Based Evaluation of the Localization Framework
123

easily set up and implemented at emergency scenes. Secondly, the computational time for
updating the location information did not took more than five seconds based on
observations in the field tests. Thirdly, the IMLE shared the same prototype with the
EASBL in the field tests, therefore, like the EASBL, the IMLE satisfies the requirements
about the size and weight of devices.
8.5 Conclusions
This chapter reports the performance of the indoor localization framework evaluated
through the field tests. The field tests were conducted in the same test bed building as the
one used in the simulation. The framework was tested under two situations, each situation
involving two fire emergency scenarios. Under situation 1, where first responders need to
set up ad-hoc networks at emergency scenes, the framework implemented the EASBL
algorithm. Respective room-level accuracies of 82.8% and 83.7%, and respective
coordinate-level accuracies of 2.12 m and 2.29 m were reported for scenario 1 and
scenario 2. Under situation 2, where existing sensing infrastructure can be used to support
the localization, the framework implemented the IMLE algorithm. Respective room-level
accuracies of 84.6% and 86.2%, and respective coordinate-level accuracies of 1.85 m and
2.07 m were reported for scenario 1 and scenario 2. Moreover, both algorithms showed
considerable robustness against losses of deployed devices that could be caused by
various hazards at emergency scenes such as fire and water. The overall performance of
the framework was promising in the field tests. It suggested that the framework has the
potential to satisfy the indoor localization requirements and support first responders with
effective access to location information.

Chapter 9: Discussions
This chapter presents further analyses and discussions of the performance of the
framework, based on the findings reported in chapters 7 and 8. Specifically, this chapter
synthesizes the results from the simulation with the results from the field tests, and then
compares the performance of the EASBL with the performance of the IMLE.
9.1 Comparison of the Simulation Results and Field Test Results
9.1.1 Performance of the EASBL Algorithm
The reported accuracies of the EASBL are summarized in Table 12. As can be seen in the
table, both in the simulation and the field tests, the performance of the EASBL was
consistent across the two scenarios, despite the variations in the size and layout of the
sensing areas. The discrepancies of the room-level accuracy were less than 0.9%, and the
discrepancies of the coordinate-level accuracy were less than 0.21 m. The accuracies
reported in the field tests were slightly lower than those reported in the simulation. The
difference could have been caused by multiple reasons, such as the existence of complex
environmental factors that interfered with the propagation of RF signals, the influence of
targets’ movements, and variations in the test subjects’ gestures of holding the
smartphones. In addition, it was observed both in the simulations and the field tests that
for a large portion of the targets that were incorrectly estimated with the EASBL at the
room level, their estimated rooms were neighboring their actual rooms (83.6% and 41.7%
in the simulation and the field tests, respectively).
Chapter 9: Discussions
125

Table 12: Localization accuracy of the EASBL
Scenarios Performance of the EASBL
Simulation Field test
Room-level
accuracy (%)
Coordinate-
level accuracy
(m)
Room-level
accuracy (%)
Coordinate-
level accuracy
(m)
Scenario 1 87.0 ± 3.6 1.78 ± 0.24 82.8 2.12
Scenario 2 87.2 ± 3.8 1.57 ± 0.22 83.7 2.29

Moreover, the measured robustness of the EASBL in the simulation was generally
consistent with the robustness measured in the field tests. The magnitude of reduction of
the room-level accuracy with increased loss of deployed nodes was quite comparable
between the simulation and the field test, as can been seen in Figure 21. Given the same
number of remaining deployed nodes, the average difference of the room-level accuracy
between the simulation and the field tests was 1.8% with slight variations. The reduction
of accuracy occurred at a lower rate in scenario 2 both in the simulation and the field tests,
due to the larger number of nodes initially deployed in the sensing area. In general, it can
be concluded that the overall performance of the EASBL was comparable between the
simulation and field tests.
Chapter 9: Discussions
126

Figure 21: Comparison of robustness between simulation and field test
9.1.2 Performance of the IMLE Algorithm
The reported accuracies of the IMLE are summarized in Table 13. Both in the simulation
and the field tests, the accuracy was consistent across the two scenarios. The
discrepancies of the room-level accuracy were less than 1.6%, and the discrepancies of
the coordinate-level accuracy were less than 0.22 m. The accuracies were expectedly
lower than those reported in the simulation, by approximately 10% at the room level and
1.2 m at the coordinate level. These differences were the likely results of various impacts
that included but were not limited to the interference of environmental factors with the
propagation of RF signals, influence of targets’ movements, and variations in test
subjects’ gestures of holding the smartphones. In addition, it was observed both in the
simulation and the field tests that for a large portion of the targets that were incorrectly
estimated with the IMLE at the room level, their estimated rooms were neighboring their
actual rooms (87.1% and 62.5% in the simulation and the field tests, respectively).

Chapter 9: Discussions
127

Table 13: Localization accuracy of the IMLE
Scenarios Performance of the IMLE
Simulation Field test
Room-level
accuracy (%)
Coordinate-
level accuracy
(m)
Room-level
accuracy (%)
Coordinate-
level accuracy
(m)
Scenario 1 95.0 ± 2.5 0.72 ± 0.10 84.6 1.85
Scenario 2 95.1 ± 2.7 0.84 ± 0.11 86.2 2.07

A comparison of the robustness observed in the field tests to that reported in the
simulation, as illustrated in Figure 22, showed that the IMLE was more robust in the
simulated environment. Given the same extent of loss of transmitters, the reduction in the
room-level accuracy was more significant (7.1% higher on average) in the real-world
environment, especially when the loss exceeded half of the total transmitters (up to 18.1%
higher). Consequently, while a minimum of 3 transmitters would suffice to ensure a
room-level accuracy of above 70% in the simulation in both scenarios, a minimum of 5
and 6 transmitters were required in scenario 1 and scenario 2, respectively, to achieve the
same level of accuracy in the field tests.
Chapter 9: Discussions
128

Figure 22: Comparison of robustness against loss of transmitters between simulation and
field tests
As for the robustness against the loss of transceivers, it was found in chapter 8.4.3 that,
unlike the simulation, the tests reported large reductions in the accuracy to an unreliable
level of below 55% when a third transceiver was removed. The discrepancy of the impact
of removing a third transceiver between the simulation and the field tests is further
illustrated in Figure 23. The figure shows that the magnitude of the accuracy reduction
was generally comparable between the simulation (blue lines) and the field tests (red
lines) in both scenario 1 (circles) and scenario 2 (triangles), except when a third
transceiver was removed.
Chapter 9: Discussions
129

Figure 23: Comparison of robustness against loss of transceivers between simulation and
field tests
9.2 Comparison between the EASBL and the IMLE Algorithms
This subchapter compares the performance of the EASBL algorithm and the IMLE
algorithm. It is important to emphasize that these two algorithms are designed with
different intentions for different situations. This inherently determines their different
performances even when used in the same building fire emergency scenarios.
The accuracies of these two algorithms both in the simulation and field tests are
summarized in Table 14. As can be seen in the table, the accuracy of the EASBL was
generally lower than that of the IMLE. In the simulation, the room-level accuracy of the
EASBL was 8.0% and 7.9% lower than that of the IMLE in scenario 1 and scenario 2,
respectively; the coordinate-level accuracy of the EASBL was 1.06 m and 0.73 m lower
than that of the IMLE in scenario 1 and scenario 2, respectively. In the field tests, the
room-level accuracy of the EASBL was 1.8% and 2.5% lower than that of the IMLE in
Chapter 9: Discussions
130

scenario 1 and scenario 2, respectively; the coordinate-level accuracy of the EASBL was
0.27 m and 0.22 m lower than that of the IMLE in scenario 1 and scenario 2, respectively.
There are two reasons that explain the relatively lower accuracy of the EASBL. First, the
accuracy was one of the two objectives the EASBL aims to optimize. By selecting
deployment plans that balance the accuracy and the deployment effort, the EASBL did
not demonstrate its full capacity in improving the accuracy. Second, the IMLE utilized
more sensing infrastructure, including all sensor nodes in the same placement utilized by
the EASBL, and a number of additional transceivers. The extended sensing infrastructure
and the resulting richer sensor data contributed to the higher accuracy yielded by the
IMLE. In addition, smaller discrepancies in the accuracy of the two algorithms were
observed in the field tests than in the simulation. The cause of such reduced discrepancies
is that the environmental factors that the simulation did not factor in, such as the
existence of metal furniture and the movement of people, had more impact on the
accuracy of the IMLE. Such stronger impact could result from the fact that the IMLE,
unlike the EASBL, relied on a mapping between signal strength and physical distances,
and the mapping was sensitive to the impact of the environmental factors.
Table 14: Comparison of the accuracy between the EASBL and the IMLE
Simulation
Scenario 1 Scenario 2
EASBL IMLE EASBL IMLE
Chapter 9: Discussions
131

Room-level accuracy (%) 87.0 ± 3.6 95.0 ± 2.5 87.2 ± 3.8 95.1 ± 2.7
Coordinate-level accuracy (m) 1.78 ± 0.24 0.72 ± 0.10 1.57 ± 0.22 0.84 ± 0.11
Field test
Scenario 1 Scenario 2
EASBL IMLE EASBL IMLE
Room-level accuracy (%) 82.8 84.6 83.7 86.2
Coordinate-level accuracy (m) 2.12 1.85 2.29 2.07

The IMLE also outperformed the EASBL in terms of the robustness. As can be seen in
Figure 24 and Figure 25, in both scenario 1 (red solid lines) and scenario 2 (blue dashed
lines), the EASBL (circles) experienced faster reductions in the room-level accuracy than
the IMLE (triangles) when the number of deployed nodes was gradually reduced. With
the minimum number of deployed nodes, i.e. 2 nodes for the EASBL and 3 for the IMLE,
the accuracy of the EASBL was always below 20%, while the accuracy of the IMLE was
always above 40%. The higher robustness of the IMLE results from its use of the MLE
method in location computation. The performance of the MLE method should not vary
significant with variances in the number of input data entries, if there are no outliers in
the input data.

Chapter 9: Discussions
132

Figure 24: Comparison of the robustness of the two algorithms (simulation)

Figure 25: Comparison of the robustness of the two algorithms (field test)
Both the EASBL and the IMLE were proven in the field tests to be deployable on scene
within the 135 s threshold. Compared to the IMLE, the EASBL required extra time
during the on-scene deployment to address the challenge of having to establish ad-hoc
sensor networks. It is critical to note that the lack of the ability in the IMLE to address
this challenge determines the IMLE is not applicable to situation 1, while the EASBL,
Chapter 9: Discussions
133

with reduced accuracy and robustness, is applicable to situation 2. The EASBL reduced
the ad-hoc network deployment effort by 32.5% in scenario 1 and 31.7% in scenario 2. In
addition, the computational time for updating the location information was comparable
across the two algorithms and in both cases far less than the 40.34 s threshold. Both
algorithms shared the same prototype in the field tests, and therefore both satisfied the
thresholds of 107.34 cm
3
and the 1.16 kg for device size and device weight, respectively.

Chapter 10: Limitations
While the localization framework has achieved promising results both in the simulation
and the field tests, it bears a number of limitations that need to be noted. First, some
caution is needed when generalizing the reported results. The framework has been tested
for two scenarios in two situations and yielded consistent performances. However, it
needs to be pointed out that the selected test bed building is not representative of all
building types, nor are the two emergency scenarios representative of all emergency
scenarios that could happen in the real world. Admittedly, the performance of the
framework may differ when implemented in other buildings, especially those with
drastically different construction types, interior spatial layouts, construction materials and
furniture. For instance, room-level localization becomes more challenging in densely
partitioned spaces filled with metal-made furniture than in open-plan spaces with few
obstructions. The performance may also differ when the nature, scope and severity of the
emergencies are different than those used in the simulation and the field tests. For
instance, fire emergencies spreading across multiple floors are more likely to damage
existing devices and leave less room for first responders to deploy ad-hoc networks, and
would challenge the scalability of the localization framework. Further implementation of
the framework in diverse test beds and emergency scenarios is necessary in order to
perform more conclusive evaluation of the framework.
Furthermore, the framework was not tested at real building emergency scenes. Setting up
a real emergency in the test bed building is prohibited by safety regulations. The
Chapter 10: Limitations
135

framework could not be tested in fire departments’ drills either due to liability and
logistical issues. Such field tests, if could be carried out in the future, would provide
valuable evidence to address the following seven critical questions regarding the
performance of the algorithms when challenged by real emergencies: (1) How accurate
and robust is the localization framework when challenged by hazards on emergency
scenes? (2) What additional constraints would hazards on emergency scenes impose on
the deployment of ad-hoc networks? (3) How reliable is the RF sensor data transmitted
and received by devices functioning under high temperature? (4) To what extent are
device losses likely to happen? (5) How reliable is the WiFi based or cellular network
based communication between smartphones and the webserver? (6) How to quickly
determine the sensing area boundary that is needed to initialize the localization
framework? (7) How to best fit the devices into the coat or backpacks worn by first
responders in order to protect the devices while avoiding intrusiveness? The answers to
these questions would further clarify the validity and applicability of the framework.
Equally important, the evaluation of the framework was based on a critical assumption
that all targets would have access to mobile nodes that can collect RF signal data and
transfer the data to a remote server. Admittedly, this assumption may not always be
satisfied in real-world situations, especially among trapped occupants. However, this
assumption would become more realistic if the framework is applied to certain types of
buildings, such as government buildings and healthcare facilities, where employers can
mandate the use of required mobile nodes. The prototype used in this thesis has proven
that smartphones can be used as the mobile nodes, avoiding intrusiveness to the
Chapter 10: Limitations
136

occupants. Implementing the framework among first responders would be easier, as first
responders have indicated in the survey the feasibility of carrying mobile nodes that do
not exceed the size and weight limits during their operations.
The framework is subject to a few other limitations. The framework needs to be
integrated into the incident command system (ICS) concept. The ICS provides a
systematic tool used for the command, control, and coordination of emergency response
operations, and is widely endorsed, sometimes mandated, by the federal agencies that
have emergency response responsibilities. The data flow and interface of the localization
framework need to be redesigned to be interoperable with the ICS, so that the localization
information can be shared by multiple agencies collaborating in emergency response
operations.
In addition, as discussed in chapter 8.3.3, the processing of geometries extracted from
BIM models when implementing the EASBL was observed in the field tests to be
prohibitively time-consuming. There are several potential solutions to address this
limitation: one can test other BIM tools that can process building geometries faster, or
extract geometries from building models and process them with a customized code, or
simply wait for the vendor to improve the computational efficiency of the BIM tool.
Lastly, both the IMLE and the EASBL rely on BIM as a source of building information.
As discussed in chapter 6.2.2, currently the accessibility of first responders to BIM is still
low.
Chapter 10: Limitations
137

Finally, the framework is based on and applicable to RF technologies only. RF
technologies, which include but are not limited to RFID, WLAN, UWB and WSN, are
rapidly evolving. The framework is essentially designed for any type of RF technologies.
Those running at frequencies compatible with onboard sensors in smartphones are
preferred, so that the implementation of the framework requires building occupants to
carry no additional devices and is hence non-intrusive. However, the framework is not
intended to be used with competing localization technologies, such as INS and PIR, that
are evaluated in chapter 5. The algorithms are designed to process RF signal data, and are
not capable of processing e.g. inertial sensor data. Admittedly, there is possibility that,
with their new development in the future, these competing technologies may outperform
the RF technologies in indoor localization. It therefore requires future research to extend
the capacities of the framework, by examining how the competing technologies can be
integrated in the framework to reduce its technology-dependence and better satisfy the
indoor localization requirements.

Chapter 11: Conclusions
The ability to locate deployed first responders and trapped occupants accurately and
quickly is of significant importance to the success of building fire emergency response
operations. This thesis examines the requirements that are important for indoor
localization at building emergency scenes. The following requirements were identified in
a survey responded by first responders nationwide: accuracy, ease of on-scene
deployment, robustness (resistance to heat, water and other physical damages),
computational speed (speed of calculating and presenting location information), and size
and weight of devices. An indoor localization framework is then introduced in this thesis
to meet these requirements. When no sensing infrastructure is available in a building, the
framework implements the EASBL algorithm, which is designed to improve room-level
localization accuracy and reduce the deployment effort of an ad-hoc sensor network, by
proposing metrics for the space division quality and deployment effort, and optimizing an
objective function that balances the tradeoff between the above two objectives. BIMs are
integrated to extract geometric information of sensing areas, support the computation of
space division quality, and provide a GUI for user interaction. Metaheuristics are
integrated to efficiently search for a solution that, although not necessarily global optima,
provides a satisfactory solution within reduced computational time compared with a
complete search. When there is existing sensing infrastructure in a building that can be
used for the collection of RF signals needed for location computation, the framework
implements the IMLE algorithm. The IMLE integrates the MLE method for estimating
Chapter 11: Conclusions
139

the parameter values of the RF signal model, and introduces an iterative process for
identifying and offsetting the impact of signal attenuations caused by walls. The
algorithm relies on a particular fitness function for evaluating location estimations that do
not converge in the iterative process, and also relies on BIM models as source of building
geometries.
The framework was evaluated both in simulation and field tests, where two building
emergency scenarios and two situations were used. In the simulation, the framework
yielded room-level accuracies of above 87.0% and coordinate-level accuracies of above
1.78 m for the EASBL, and room-level accuracies of above 95.0% and coordinate-level
accuracies of above 0.84 m for the IMLE. In the field tests, the framework yielded room-
level accuracies of above 82.8% and coordinate-level accuracies of above 2.29 m for the
EASBL, and room-level accuracies of above 84.6% and coordinate-level accuracies of
above 2.07 m for the IMLE. The ease of on-scene deployment of the framework, when
establishment of ad-hoc networks is needed and therefore the EASBL is implemented, is
addressed by the design of the EASBL that takes reducing deployment effort as one of
the algorithm’s two objectives, the other one being improving the accuracy. The
framework yielded promising robustness both in the simulation and the field tests, by
ensuring satisfactory accuracies upon partial loss of the sensing network. Moreover, the
framework has been proven capable of meeting the requirements about computational
speed and the size and weight of devices in the field tests.
There are several limitations, such as limited representativeness of the test bed and
emergency scenarios, simplicity of real building emergency situations, requirement on
Chapter 11: Conclusions
140

the use of mobile nodes, and incompatibility with competing indoor localization
technologies. Addressing these limitations, although outside the scope of this thesis, is
expected to further improve the performance of the framework, and advance this line of
research in the future.

References
[1] MJ Karter, Fire Loss in the United States During 2012, National Fire Protection
Association, Fire Analysis and Research Division, Quincy, MA, 2013.
[2] E Brouwer, Trainer's Corner: What Are Your Rules for Calling Mayday?
http://www.firefightingincanada.com/content/view/1274/213/. Last accessed June 1, 2012.
[3] USFA, Fire-Related Firefighter Injuries Reported to NFIRS, Topical Fire Report
Series. U.S. Fire Administration, Emmitsburg, Maryland, 2011.
[4] RF Fahy, U.S. Fire Service Fatalities in Structure Fires, 1977-2009, National Fire
Protection Association, Quincy, MA, 2010.
[5] USFA, Firefighter Fatalities in the United States in 2011, U.S. Fire Administration,
Emmitsburg, Maryland, 2012.
[6] Carpentersville Fire Department, Fire Department Procedures,
http://vil.carpentersville.il.us/Services/Fire/Procedures.asp. Last accessed June 26 2012.
[7] Headquaters, Firefighting and Rescue Procedures in Theaters of Operations,
Department of the Army, TM 5-315 Washington D.C., 1971.
[8] Fire Service Resources Network. Primary and Secondary Search, Weekly Drills. 6
(2006).
References
142

[9] N Li, Z Yang, A Ghahramani, B Becerik-Gerber, L Soibelman. Situational Awareness
for Supporting Building Fire Emergency Response: Information Needs, Information
Sources, and Implementation Requirements, Fire Saf.J. 63 (2014) 17-28.
[10] EG Hinkelman, S Putzi, Dictionary of International Trade: Handbook of the Global
Trade Community, World Trade Press, 2005.
[11] G Cybenko, B Brewington, The Foundation of Information Push and Pull, in:
O'Leary D (Ed.), Mathematics of Information Coding, Extraction and Distribution,
Verlag: Springer, 1998, pp. 9-30.
[12] X Fan, J Yen, RA Volz. A Theoretical Framework on Proactive Information
Exchange in Agent Teamwork, Artif.Intell. 169 (2005) 23-97.
[13] R Weitzman, G Rabinowitz. Sensitivity of 'Push' and 'Pull' Strategies to Information
Updating Rate, Int J Prod Res. 41 (2003) 2057-2074.
[14] ML Spearman, DL Woodruff, WJ Hopp. CONWIP. A Pull Alternative to Kanban,
Int J Prod Res. 28 (1990) 879-894.
[15] D Cummings, Push Vs. Pull - The Battle for the Best CMS - SitePoint,
http://www.sitepoint.com/push-pull-best-cms/. Last accessed May 3, 2012.
[16] M Muethel, Mixed Methods Application in Trust Research: Simultaneous Hybrid
Data Collection in Cross-Cultural Settings Using the Board-Game Method, in: Lyon HF
(Ed.), Handbook of Research Methodology on Trust, Edward Elgar Pub, 2012.
References
143

[17] JN Follin, PS Fischbeck, Trade-Offs Among Environmental, Human Health and
Quality-of-Life Impacts, in: Improving Regulation: Cases in Environment, Health, and
Safety, Routledge, 2001, pp. 186-207.
[18] S Kiziltas, B Akinci, C Gonzalez. Comparison of Experienced and Novice Cost
Estimator Behaviors in Information Pull and Push Methods, Canadian Journal of Civil
Engineering. 37 (2010) 290-301.
[19] S Kiziltas, B Akinci. Contextual Information Requirements of Cost Estimators from
Past Construction Projects, J.Constr.Eng.Manage. 135 (2009) 841-852.
[20] S Kiziltas, B Akinci, C Gonzalez. Understanding Differences in Information Needs
of Expert and Novice Estimators from Construction Project Histories, Proc. of
Construction Research Congress 2007, Grand Bahama Island, May 6-8, 2007.
[21] G Atasoy, Visualizing and Interacting with Construction Project Performance
Information, Doctoral Dissertation. Carnegie Mellon University, Pittsburgh, PA, 2013.
[22] DG Holmberg, ST Bushby, KA Reed, Workshop to Define Information Needed by
Emergency Responders during Building Emergencies, NISTIR 7314. U.S Department of
Commerce, National Institute of Standard and Technology, NISTIR 7193, Gaithersburg,
Maryland, 2005.
[23] DG Holmberg, WD Davis, SJ Treado, KA Reed, Building Tactical Information
System for Public Safety Officials, NISTIR 7314. U.S Department of Commerce,
National Institute of Standard and Technology, NISTIR 7314, Gaithersburg, MD, 2006.
References
144

[24] N Li, B Becerik-Gerber, B Krishnamachari, L Soibelman. A BIM Centered Indoor
Localization Algorithm to Support Building Fire Emergency Response Operations,
Autom.Constr. 42 (2014) 78-89.
[25] H Akcan, C Evrendilek. GPS-Free Directional Localization via Dual Wireless
Radios, Comput.Commun. 35 (2012) 1151-1163.
[26] A Lo, L Xia, I Niemegeers, T Bauge, M Russell, D Harmer. EUROPCOM - An
Ultra-Wideband (UWB)-Based Ad Hoc Network for Emergency Applications, Proc. of
VTC/Spring - 2008 IEEE 67th Vehicular Technology Conference, (2008), Singapore, 11-
14 May 2008.
[27] J Rantakokko, J Rydell, P Stromback, P Handel, J Callmer, D Tornqvist, et al.
Accurate and Reliable Soldier and First Responder Indoor Positioning: Multisensor
Systems and Cooperative Localization, IEEE Wireless Communications. 18 (2011) 10-18.
[28] A Chandra-Sekaran, P Weisser, K Muller-Glaser, C Kunze. A Comparison of
Bayesian Filter Based Approaches for Patient Localization During Emergency Response
to Crisis, Proc. of 2009 Third International Conference on Sensor Technologies and
Applications (SENSORCOMM), (2009), Piscataway, NJ, 18-23 June 2009.
[29] A Chandra-Sekaran, G Stefansson, C Kunze, K Muller-Glaser, P Weisser. A Range-
Based Monte Carlo Patient Localization During Emergency Response to Crisis, Proc. of
2009 5th Advanced International Conference on Telecommunications, AICT 2009,
Venice, Italy, 24-28 May, 2009.
References
145

[30] J Duckworth, D Cyganski, S Makarov, W Michalson, J Orr, V Amendolare, et al.
WPI precision personnel locator system - Evaluation by first responders, Proc. of 20th
International Technical Meeting of the Satellite Division of The Institute of Navigation
2007 ION GNSS 2007, Fort Worth, TX, 25-28 September, 2007.
[31] D Cyganski, J Duckworth, S Makarov, W Michalson, J Orr, V Amendolare, et al.
WPI Precision Personnel Locator system, Proc. of Institute of Navigation National
Technical Meeting, NTM 2007, San Diego, CA, 22-24 January, 2007.
[32] B Woodacre, D Cyganski, RJ Duckworth, V Amendolare. WPI precision personnel
locator system: Antenna Geometry Estimation Using a Robust Multilateralization
Technique, Proc. of Institute of Navigation - International Technical Meeting, ITM 2009,
Anaheim, CA, 26-28 January, 2009.
[33] A Cavanaugh, M Lowe, D Cyganski, RJ Duckworth. WPI Precision Personnel
Location System: Rapid Deployment Antenna System and Sensor Fusion for 3D
Precision Location, Proc. of Institute of Navigation - International Technical Meeting
2010, ITM 2010, San Diego, CA, 25-27 January, 2010.
[34] HM Khoury, VR Kamat. Evaluation of Position Tracking Technologies for User
Localization in Indoor Construction Environments, Autom.Constr. 18 (2009) 444-457.
[35] JR Guerrieri, MH Francis, PF Wilson, T Kos, LE Miller, NP Bryner, et al. RFID-
Assisted Indoor Localization and Communication for First Responders, Proc. of First
References
146

European Conference on Antennas and Propagation Conference, Nice, France, 6-10 Nov.
2006.
[36] U Ruppel, KM Stubbe, U Zwinger. Indoor Navigation Integration Platform for
firefighting purposes, Proc. of 2010 International Conference on Indoor Positioning and
Indoor Navigation (IPIN 2010), Piscataway, NJ, 15-17 Sept. 2010.
[37] AO Kaya, L Greenstein, D Chizhik, R Valenzuela, N Moayeri. Emitter Localization
and Visualization (ELVIS): A Backward Ray Tracing Algorithm for Locating Emitters,
Proc. of the 41st Annual Conference on Information Sciences and Systems, Piscataway,
NJ, 14-16 March 2007.
[38] J Mapar, Tracking Emergency Responders in Challenging Environments,
http://spie.org/x39740.xml?ArticleID=x39740. Last accessed April 6, 2012.
[39] Exit Technologies, http://www.exit-technologies.com/draeger.php. Last accessed
April 6, 2012.
[40] Y Zhao, Y Liu, LM Ni. VIRE: Active RFID-Based Localization Using Virtual
Reference Elimination, Proc. of 2007 International Conference on Parallel Processing,
Piscataway, NJ, 10-14 Sept. 2007.
[41] S Chumkamon, P Tuvaphanthaphiphat, P Keeratiwintakorn. A Blind Navigation
System Using RFID for Indoor Environments, Proc. of 5th International Conference on
Electrical Engineering/Electronics, Computer, Telecommunications and Information
Technology, ECTI-CON 2008, Krabi, Thailand, 14-17 May, 2008.
References
147

[42] J Nord, K Synnes, P Parnes. An Architecture for Location Aware Applications, Proc.
of the 35th Annual Hawaii International Conference on System Sciences, Los Alamitos,
CA, 7-10 Jan. 2002.
[43] K Eustice, V Ramakrishna, N Nguyen, P Reiher. The smart party: A Personalized
Location-Aware Multimedia Experience, Proc. of the 5th IEEE Consumer
Communications and Networking Conference, Las Vegas, NV, 10-12 Jan. 2008.
[44] AR Jimenez, F Seco, C Prieto, J Guevara. A Comparison of Pedestrian Dead-
Reckoning Algorithms Using a Low-Cost MEMS IMU, Proc. of WISP 2009, Piscataway,
NJ, 26-28 Aug. 2009.
[45] R Zhang, LM Reindl. Inertial Localization System Using Unscented Kalman Filter
for 3D Positioning, Proc. of 4th International Congress on Image and Signal Processing
(CISP 2011), Piscataway, NJ, 15-17 Oct. 2011.
[46] B Krach, P Robertson. Integration of Foot-Mounted Inertial Sensors Into a Bayesian
Location Estimation Framework, Proc. of 5th Workshop on Positioning, Navigation and
Communication 2008, WPNC'08, Hannover, Germany, 27 March, 2008.
[47] K Yamanaka, M Kanbara, N Yokoya. Localization of Walking or Running User
with Wearable 3D Position Sensor, Proc. of 17th International Conference on Artificial
Reality and Telexistence, ICAT 2007, Los Alamitos, CA, 28-30 Nov. 2007.
References
148

[48] A Soloviev, TJ Dickman. Deeply Integrated GPS for Indoor Navigation, Proc. of
2010 International Conference on Indoor Positioning and Indoor Navigation (IPIN 2010),
Piscataway, NJ, 15-17 Sept. 2010.
[49] M Akula, S Dong, VR Kamat, L Ojeda, A Borrell, J Borenstein. Integration of
Infrastructure Based Positioning Systems and Inertial Navigation for Ubiquitous Context-
Aware Engineering Applications, Advanced Engineering Informatics. 25 (2011) 640-655.
[50] T Lee, J Shirr, D Cho. Position Estimation for Mobile Robot Using In-Plane 3-Axis
IMU and Active Beacon, Proc. of IEEE International Symposium on Industrial
Electronics, IEEE ISIE 2009, Seoul, Korea, 5-8 July, 2009.
[51] AR Jimenez Ruiz, F Seco Granja, JC Prieto Honorato, JI Guevara Rosas. Accurate
Pedestrian Indoor Navigation by Tightly Coupling Foot-Mounted IMU and RFID
Measurements, IEEE Transactions on Instrumentation and Measurement. 61 (2012) 178-
189.
[52] A Colombo, D Fontanelli, D Macii, L Palopoli. A Wearable Embedded Inertial
Platform with Wireless Connectivity for Indoor Position Tracking, Proc. of 2011 IEEE
International Instrumentation and Measurement Technology Conference (I2MTC 2011),
Piscataway, NJ, 10-12 May 2011.
[53] JA Hesch, FM Mirzaei, GL Mariottini, SI Roumeliotis. A Laser-Aided Inertial
Navigation System (L-INS) for Human Localization in Unknown Indoor Environments,
References
149

Proc. of 2010 IEEE International Conference on Robotics and Automation, ICRA 2010,
Anchorage, AK, 3-7 May, 2010.
[54] P Davidson, J Collin, J Takala. Application of Particle Filters for Indoor Positioning
Using Floor Plans, Proc. of 2010 Ubiquitous Positioning Indoor Navigation and Location
Based Service, UPINLBS 2010, Helsinki, Finland, 14-15 October, 2010.
[55] O Woodman, R Harle. Pedestrian Localisation for Indoor Environments, Proc. of
10th International Conference on Ubiquitous Computing, UbiComp 2008, Seoul, Korea,
21-24 September, 2008.
[56] U Walder, T Bernoulli. Context-Adaptive Algorithms to Improve Indoor Positioning
with Inertial Sensors, Proc. of 2010 International Conference on Indoor Positioning and
Indoor Navigation (IPIN 2010), Piscataway, NJ, 15-17 Sept. 2010.
[57] G Glanzer, T Bernoulli, T Wieflecker, U Walder. Semi-Autonomous Indoor
Positioning Using MEMS-Based Inertial Measurement Units and Building Information,
Proc. of 6th Workshop on Positioning, Navigation and Communication, WPNC 2009,
Hannover, Germany, 19 March 2009.
[58] vF Diggelen. Indoor GPS Theory Implementation, Proc. of IEEE Position Location
and Navigation Symposium, Piscataway, NJ, 15-18 April 2002.
[59] F Dovis, R Lesca, D Margaria, G Boiero, G Ghinamo. An Assisted High-Sensitivity
Acquisition Technique for GPS Indoor Positioning, Proc. of 2008 IEEE/ION Position,
Location and Navigation Symposium - PLANS 2008, Piscataway, NJ, 05 May 2008.
References
150

[60] Y Liu, S Tian. Research of Indoor GPS Signals Acquisition Algorithm, Proc. of 4th
International Conference on Wireless Communications, Networking and Mobile
Computing (WiCOM), Piscataway, NJ, 12-14 Oct. 2008.
[61] O Bayrak, T Goze, M Barut, MO Sunay. Analysis of SUPL A-GPS (Secure User
Plane Location) in Indoor Areas, Proc. of 2008 IEEE Region 8 International Conference
on Computational Technologies in Electrical and Electronics Engineering, SIBIRCON
2008, Novosibirsk, Russia, 21-25 July 2008.
[62] K Ozsoy, A Bozkurt, I Tekin. 2D Indoor Positioning System Using GPS Signals,
Proc. of 2010 International Conference on Indoor Positioning and Indoor Navigation
(IPIN 2010), Piscataway, NJ, 15-17 Sept. 2010.
[63] A- Anwar, G Ioannis, FN Pavlidou. Evaluation of Indoor Location Based on
Combination of AGPS/ HSGPS, Proc. of 3rd International Symposium on Wireless
Pervasive Computing (ISWPC 2008), Piscataway, NJ, 7-9 May 2008.
[64] A- Anwar, G Ioannis, FN Pavlidou. Indoor Location Tracking Using AGPS and
Kalman Filter, Proc. of 2009 6th Workshop on Positioning, Navigation and
Communication (WPNC'09), Piscataway, NJ, 19 March 2009.
[65] T Aytac, B Barshan. Target Differentiation and Localization Using Infrared Sensors,
Proc. of 19th Congress of the International Commisssion for Optics Optics for the
Quality of Life, Firenze, Italy, 25-30 Aug. 2002.
References
151

[66] T Aytac, B Barshan. Differentiation and Localization of Targets Using Infrared
Sensors, Opt.Commun. 210 (2002) 25-35.
[67] S Hijikata, K Terabayashi, K Umeda. A Simple Indoor Self-Localization System
Using Infrared LEDs, Proc. of 2009 Sixth International Conference on Networked
Sensing Systems (INSS 2009), Piscataway, NJ, 17-19 June 2009.
[68] E Brassart, C Pegard, M Mouaddib. Localization Using Infrared Beacons, Robotica.
18 (2000) 153-61.
[69] RC Luo, O Chen, HL Pei. Indoor Robot/Human Localization Using Dynamic
Triangulation and Wireless Pyroelectric Infrared Sensory Fusion Approaches, Proc. of
2012 IEEE International Conference on Robotics and Automation (ICRA), Piscataway,
NJ, 14-18 May 2012.
[70] S Tao, M Kudo, H Nonaka, J Toyama. Recording the Activities of Daily Living
Based on Person Localization Using an Infrared Ceiling Sensor Network, Proc. of 2011
IEEE International Conference on Granular Computing, GrC 2011, Kaohsiung, Taiwan,
8-10 Nov. 2011.
[71] N Petrellis, N Konofaos, GP Alexiou. Target Localization Utilizing the Success Rate
in Infrared Pattern Recognition, IEEE Sensors Journal. 6 (2006) 1355-64.
[72] J Kemper, H Linde. Challenges of Passive Infrared Indoor Localization, Proc. of 5th
Workshop on Positioning, Navigation and Communication (WPNC '08), Hannover,
Germany, 27 March 2008.
References
152

[73] D Hauschildt, N Kirchhof. Improving Indoor Position Estimation by Combining
Active TDOA Ultrasound and Passive Thermal Infrared Localization, Proc. of 8th
Workshop on Positioning Navigation and Communication 2011, WPNC 2011, Dresden,
Germany, 7-8 Apr. 2011.
[74] D Hauschildt, N Kirchhof. Advances in Thermal Infrared Localization: Challenges
and Solutions, Proc. of 2010 International Conference on Indoor Positioning and Indoor
Navigation (IPIN 2010), Piscataway, NJ, 15-17 Sept. 2010.
[75] J Hightower, G Borriello, R Want, SpotON: An Indoor 3D Location Sensing
Technology Based on RF Signal Strength, Department of Computer Science and
Engineering, University of Washington, Seattle, WA, 2000, 1-16.
[76] K Yu, C Liao, M Lee, H Lin. Design and Implementation of a RFID Based Real-
Time Location-Aware System in Clean Room, Proc. of 2009 IEEE International
Symposium on Parallel and Distributed Processing with Applications, ISPA 2009,
Chengdu, China, 9-12 Aug. 2009.
[77] X Luo, WJ O'Brien, CL Julien. Comparative evaluation of Received Signal-Strength
Index (RSSI) Based Indoor Localization Techniques for Construction Jobsites, Advanced
Engineering Informatics. 25 (2) (2010) 355–363.
[78] LM Ni, Y Liu, CL Yiu, AP Patil. LANDMARC: Indoor Location Sensing Using
Active RFID, Wireless Networks. 10 (2004) 701-10.
References
153

[79] S Polito, D Biondo, A Iera, M Mattei, A Molinaro. Performance Evaluation of
Active RFID Location Systems Based on RF Power Measures, Proc. of 18th Annual
IEEE International Symposium on Personal, Indoor and Mobile Radio Communications,
PIMRC'07, Athens, Greece, 3-7 Sep. 2007.
[80] T Zhang, Z Chen, Y Ouyang, J Hao, Z Xiong. An Improved RFID-Based Locating
Algorithm by Eliminating Diversity of Active Tags for Indoor Environment, Comput.J.
52 (2009) 902-9.
[81] PW Hsu, TH Lin, HH Chang, YT Chen, CY Yen, YJ Tseng, et al. Practicability
Study on the Improvement of the Indoor Location Tracking Accuracy with Active RFID,
Proc. of CMC 2009, Piscataway, NJ, 6-8 Jan. 2009.
[82] N Li, S Li, B Becerik-Gerber, G Calis. Deployment Strategies and Performance
Evaluation of a Virtual-Tag-Enabled Indoor Location Sensing Approach, Journal of
Computing in Civil Engineering. 26 (2012) 574-583.
[83] K Sue, C Tsai, M Lin. FLEXOR: A Flexible Localization Scheme Based on RFID,
International Conference on Information Networking, Proc. of ICOIN 2006, Sendai,
Japan, 16-19 Jan. 2006.
[84] Y Huang, Z Lui, G Ling. An Improved Bayesian-Based RFID Indoor Location
Algorithm, Proc. of 2008 International Conference on Computer Science and Software
Engineering (CSSE 2008), Piscataway, NJ, 12-14 Dec. 2008.
References
154

[85] RdA Silva, PAdS Goncalves. Enhancing the Efficiency of Active RFID-Based
Indoor Location Systems, Proc. of 2009 IEEE Wireless Communications and Networking
Conference, WCNC 2009, Budapest, Hungary, 5-8 Apr. 2009.
[86] MA Khan, VK Antiwal. Location Estimation Technique Using Extended 3-D
LANDMARC Algorithm for Passive RFID Tag, Proc. of IACC 2009, (2009), Piscataway,
NJ, 6-7 March 2009.
[87] G Jin, X Lu, M Park. An Indoor Localization Mechanism Using Active RFID Tag,
Proc. of IEEE International Conference on Sensor Networks, Ubiquitous, and
Trustworthy Computing, Taichung, Taiwan, 5-7 June 2006.
[88] X Wang, X Jiang, Y Liu. An Enhanced Approach of Indoor Location Sensing Using
Active RFID, Proc. of 2009 WASE International Conference on Information Engineering
(ICIE), Piscataway, NJ, 10-11 July 2009.
[89] W Li, J Wu, D Wang. A Novel Indoor Positioning Method Based on Key Reference
RFID Tags, Proc. of 2009 IEEE Youth Conference on Information, Computing and
Telecommunication (YC-ICT 2009), Piscataway, NJ, 20-21 Sept. 2009.
[90] Y Huang, S Lv, Z Liu, W Jun, S Jun. The Topology Analysis of Reference Tags of
RFID Indoor Location System, Proc. of 3rd IEEE International Conference on Digital
Ecosystems and Technologies (DEST), Piscataway, NJ, 1-3 June 2009.
References
155

[91] Z Zhen, Q Jia, C Song, X Guan. An Indoor Localization Algorithm for Lighting
Control Using RFID, Proc. of 2008 IEEE Energy 2030 Conference, ENERGY 2008,
Atlanta, GA, 17-18 Nov. 2008.
[92] U Rueppel, KM Stuebbe. BIM-Based Indoor-Emergency-Navigation-System for
Complex Buildings, Tsinghua Science & Technology. 13 (2008) 362-367.
[93] A Pradhan, E Ergen, B Akinci. Technological Assessment of Radio Frequency
Identification Technology for Indoor Localization, J.Comp.in Civ.Engrg. 23 (2009) 230-
238.
[94] U Grossmann, M Schauch, S Hakobyan. RSSI Based WLAN Indoor Positioning
with Personal Digital Assistants, Proc. of 4th IEEE Workshop on Intelligent Data
Acquisition and Advanced Computing Systems: Technology and Applications, IDAACS,
Dortmund, Germany, 6-8 September 2007.
[95] S Gansemer, S Hakobyan, S Puschel, U Gromann. 3D WLAN Indoor Positioning in
Multi-Story Buildings, Proc. of 5th IEEE International Workshop on Intelligent Data
Acquisition and Advanced Computing Systems: Technology and Applications,
IDAACS'2009, Rende, Italy, 21-23 September 2009.
[96] Shih-Hau Fang, T Lin. Principal Component Localization in Indoor WLAN
Environments, IEEE Transactions on Mobile Computing. 11 (2012) 100-10.
[97] Shih-Hau Fang, Tsung-Nan Lin. Accurate WLAN Indoor Localization Based on
RSS, Fluctuations Modeling, Proc. of WISP 2009, Piscataway, NJ, 26-28 Aug. 2009.
References
156

[98] H Schmitzberger. Autonomous WLAN Sensors for Ad Hoc Indoor Localization,
Proc. of 13th International Conference, Piscataway, NJ, 6-11 Feb. 2011.
[99] L Peng, J Han, W Meng, J Liu. Research on Radio-Map Construction in Indoor
WLAN Positioning System, Proc. of First International Conference on Pervasive
Computing, Signal Processing and Applications (PCSPA 2010), Piscataway, NJ, 17-19
Sept. 2010.
[100] A del Corte-Valiente, J Gomez-Pulido, O Gutierrez-Blanco. Efficient Techniques
and Algorithms for Improving Indoor Localization Precision on WLAN Networks
Applications, International Journal of Communications, Networks and System Sciences.
2 (2009) 645-51.
[101] K El-Kafrawy, M Youssef, A El-Keyi, A Naguib. Propagation Modeling for
Accurate Indoor WLAN RSS-based Localization, Proc. of 2010 IEEE 72
nd
Vehicular
Technology Conference Fall, Piscataway, NJ, 6-9 Sept. 2010.
[102] H Wang, L Ma, Y Xu, Z Deng. Dynamic Radio Map Construction for WLAN
Indoor Location, Proc. of 2011 International Conference on Intelligent Human-Machine
Systems and Cybernetics, Piscataway, NJ, 26-27 Aug. 2011.
[103] Shih-Hau Fang, Tsung-Nan Lin, Kun-Chou Lee. A Novel Algorithm for Multipath
Fingerprinting in Indoor WLAN Environments, IEEE Transactions on Wireless
Communications. 7 (2008) 3579-88.
References
157

[104] L Limin, M Lin, X Yubin, W Jiayin. Application of Multi-Cluster-Center Based
Filtering in WLAN Indoor Positioning, Proc. of 6th International ICST Conference on
Communications and Networking in China (CHINACOM 2011), Piscataway, NJ, 17-19
Aug. 2011.
[105] L Arya, SC Sharma, M Pant. Coverage of access points using Particle Swarm
optimization in indoor WLAN, Information Processing and Management, Proc. of
International Conference on Recent Trends in Business Administration and Information
Processing, BAIP 2010, Trivandrum, India, 26-27 March 2010.
[106] O Baala, Y Zheng, A Caminada. The Impact of AP Placement in WLAN-Based
Indoor Positioning System, Proc. of 8th International Conference on Networks, ICN 2009,
Gosier, Guadaloupe, 1-6 March 2009.
[107] M Bocquet, C Loyez, A Benlarbi-Delai. Using Enhanced-TDOA Measurement for
Indoor Positioning, IEEE Microwave and Wireless Components Letters. 15 (2005) 612-
14.
[108] R Ye, S Redfield, H Liu. High-Precision Indoor UWB Localization: Technical
Challenges and Method, Proc. of 2010 IEEE International Conference on Ultra-
Wideband (ICUWB 2010), Piscataway, NJ, 20-23 Sept. 2010.
[109] H Xiong, H Song, Z Lai, J Zhang, K Yi. A Novel Indoor Localization Scheme,
Proc. of 12th IEEE International Conference on Communication Technology (ICCT
2010), Piscataway, NJ, 11-14 Nov. 2010.
References
158

[110] H Luecken, A Wittneben. Low Complexity Positioning System for Indoor
Multipath Environments, Proc. of 2010 IEEE International Conference on
Communications, ICC 2010, Cape Town, South Africa, 23-27 May 2010.
[111] P Meissner, D Arnitz, T Gigl, K Witrisal. Analysis of an Indoor UWB Channel for
Multipath-Aided Localization, Proc. of 2011 IEEE International Conference on Ultra-
Wideband (ICUWB 2011), Piscataway, NJ, 14-16 Sept. 2011.
[112] Z Li, W Dehaene, G Gielen. A 3-tier UWB-Based Indoor Localization Scheme for
Ultra-Low-Power Sensor Nodes, Proc. of 2007 IEEE International Conference on Signal
Processing and Communications, Piscataway, NJ, 24-27 Nov. 2007.
[113] SJ Ingram, D Harmer, M Quinlan. UltraWideBand Indoor Positioning Systems and
Their Use in Emergencies, Proc. of Position Location and Navigation Symposium,
Piscataway, NJ, 26-29 April 2004.
[114] MJ Kuhn, J Turnmire, MR Mahfouz, AE Fathy. Adaptive Leading-Edge Detection
in UWB Indoor Localization, Proc. of 2010 IEEE Radio and Wireless Symposium, RWW
2010, New Orleans, LA, 10-14 Jan. 2010.
[115] MR Mahfouz, C Zhang, BC Merkl, MJ Kuhn, AE Fathy. Investigation of High-
Accuracy Indoor 3-D Positioning Using UWB Technology, IEEE Trans.Microwave
Theory Tech. 56 (2008) 1316-1330.
[116] MR Mahfouz, MJ Kuhn, Y Wang, J Turnmire, AE Fathy. Towards Sub-Millimeter
Accuracy in UWB Positioning for Indoor Medical Environments, Proc. of 2011 IEEE
References
159

Topical Conference on Biomedical Wireless Technologies, Networks, and Sensing
Systems (BioWireleSS), Piscataway, NJ, 16-19 Jan. 2011.
[117] M Kuhn, C Zhang, B Merkl, D Yang, Y Wang, M Mahfouz, et al. High Accuracy
UWB Localization in Dense Indoor Environments, Proc. of 2008 IEEE International
Conference on Ultra-Wideband (ICUWB), Piscataway, NJ, 10-12 Sept. 2008.
[118] H Ahn, Sang-Burm Rhee. Simulation of a RSSI-based Indoor Localization System
Using Wireless Sensor Network, Proc. of 5th International Conference on Ubiquitous
Information Technologies & Applications (CUTE 2010), Piscataway, NJ, 16-18 Dec.
2010.
[119] P Cherntanomwong, DJ Suroso. Indoor Localization System Using Wireless Sensor
Networks for Stationary and Moving Target, Proc. of 8th International Conference on
Information, Communications & Signal Processing (ICICS 2011), Piscataway, NJ, 13-16
Dec. 2011.
[120] G Zanca, F Zorzi, A Zanella, M Zorzi. Experimental Comparison of RSSI-Based
Localization Algorithms for Indoor Wireless Sensor Networks, Proc. of 3rd Workshop on
Real-World Wireless Sensor Networks, REALWSN 2008, Glasgow, United kingdom, 1
April 2008.
[121] R Priwgharm, P Chemtanomwong. A Comparative Study on Indoor Localization
Based on RSSI Measurement in Wireless Sensor Network, Proc. of 2011 Eighth
References
160

International Joint Conference on Computer Science and Software Engineering (JCSSE
2011), Piscataway, NJ, 11-13 May 2011.
[122] K Yedavalli, B Krishnamachari. Sequence-Based Localization in Wireless Sensor
Networks, IEEE Transactions on Mobile Computing. 7 (2008) 81-94.
[123] K Yedavalli, B Krishnamachari, S Ravulat, B Srinivasan. Ecolocation: A Sequence
Based Technique for RF Localization in Wireless Sensor Networks, Proc. of 4th
International Symposium on Information Processing in Sensor Networks, IPSN 2005, Los
Angeles, CA, 25-27 April 2005.
[124] Jehn-Ruey Jiang, Chih-Ming Lin, Yi-Jia Hsu. Localization with Rotatable
Directional Antennas for Wireless Sensor Networks, Proc. of 39th International
Conference on Parallel Processing Workshops (ICPPW), Piscataway, NJ, 13-16 Sept.
2010.
[125] B Yang, J Yang, J Xu, D Yang. Area Localization Algorithm for Mobile Nodes in
Wireless Sensor Networks Based on Support Vector Machines, Proc. of 3rd International
Conference on Mobile Ad-hoc and Sensor Networks, MSN 2007, Beijing, China, 12-14
Dec. 2007.
[126] V Kaseva, TD Hamalainen, M Hannikainen. Range-Free Algorithm for Energy-
Efficient Indoor Localization in Wireless Sensor Networks, Proc. of 2011 Conference on
Design and Architectures for Signal and Image Processing, Piscataway, NJ, 2-4 Nov.
2011.
References
161

[127] X Kuai, K Yang, S Fu, R Zheng, G Yang. Simultaneous Localization and Mapping
(SLAM) for Indoor Autonomous Mobile Robot Navigation in Wireless Sensor Networks,
Proc. of 2010 International Conference on Networking, Sensing and Control (ICNSC
2010), Piscataway, NJ, 10-12 April 2010.
[128] DJ Suroso, P Cherntanomwong, P Sooraksa, J Takada. Fingerprint-Based
Technique for Indoor Localization in Wireless Sensor Networks Using Fuzzy C-Means
Clustering Algorithm, Proc. of 2011 International Symposium on Intelligent Signal
Processing and Communications Systems (ISPACS 2011), Piscataway, NJ, 7-9 Dec. 2011.
[129] L Gogolak, S Pletl, D Kukolj. Indoor Fingerprint Localization in WSN
Environment Based on Neural Network, Proc. of IEEE 9th International Symposium on
Intelligent Systems and Informatics (SISY 2011), Piscataway, NJ, 8-10 Sept. 2011.
[130] T Wu, I- Liao, W Lee, G Liaw, J Ding, C Wu. Enhancing Indoor Localization
Accuracy of Sensor-Based by Advance Genetic Algorithms, Proc. of 6th International
Wireless Communications and Mobile Computing Conference, IWCMC 2010, Caen,
France, 28 June-2 July 2010.
[131] M Dominguez-Duran, D Claros, C Urdiales, F Coslado, F Sandoval. Dynamic
Calibration and Zero Configuration Positioning System for WSN, Proc. of MELECON
2008 - 2008 IEEE Mediterranean Electrotechnical Conference, Ajaccio, France, 5-7 May
2008.
References
162

[132] Yu-Chi Chen, Jyh-Ching Juang. Outlier-Detection-Based Indoor Localization
System for Wireless Sensor Networks, International Journal of Navigation and
Observation. (2012).
[133] SA Mitilineos, JN Goufas, OE Segou, SCA Thomopoulos. WAX-ROOM: An
Indoor WSN-Based Localization Platform, Proc. of Signal Processing, Sensor Fusion,
and Target Recognition XIX, Piscataway, NJ, 5 April 2010.
[134] Z Xiong, F Sottile, MA Caceres, MA Spirito, R Garello. Hybrid WSN-RFID
Cooperative Positioning Based on Extended Kalman Filter, Proc. of IEEE-APS Topical
Conference on Antennas and Propagation in Wireless Communications, Piscataway, NJ,
12-16 Sept. 2011.
[135] Z Xiong, F Sottile, MA Spirito, R Garello. Hybrid Indoor Positioning Approaches
Based on WSN and RFID, Proc. of 4th IFIP International Conference on New
Technologies, Mobility and Security (NTMS 2011), Piscataway, NJ, 7-10 Feb. 2011.
[136] D Zhang, Y Yang, D Cheng, S Liu, LM Ni. COCKTAIL: An RF-Based Hybrid
Approach for Indoor Localization, Proc. of ICC 2010 - 2010 IEEE International
Conference on Communications, Piscataway, NJ, 23-27 May 2010.
[137] A Savvides, C- Han, MB Strivastava. Dynamic Fine-Grained Localization in Ad-
Hoc Networks of Sensors, Proc. of 7th Annual International Conference on Mobile
Computing and Networking, Rome, Italy, 16-21 July 2001.
References
163

[138] C Eastman, P Teicholz, R Sacks, K Liston, BIM Handbook: A Guide to Building
Information Modeling for Owners, Managers, Designers, Engineers, and Contractors,
illustrated ed., John Wiley and Sons 2008.
[139] B Becerik-Gerber, F Jazizadeh, N Li, G Calis. Application Areas and Data
Requirements for BIM-Enabled Facilities Management, J.Constr.Eng.Manage. 138 (2012)
431-442.
[140] J Pearl, Heuristics: Intelligent Search Strategies for Computer Problem Solving,
Addison-Wesley, New York, 1983.
[141] DT Pham, D Karaboga, Intelligent Optimisation Techniques: Genetic Algorithms,
Tabu Search, Simulated Annealing and Neural Networks, 1st ed., Springer 2000.
[142] MarketsandMarkets, Radio Frequency Components (RFC) Market for Consumer
Electronics – Global Forecast & Analysis (2012 – 2017), MarketsandMarkets, SE 1180
Dallas, TX, 2012.
[143] JM Keenan, AJ Motley. Radio Coverage in Buildings, British Telecom Technology
Journal. 8 (1990) 19-24.
[144] TS Rappaport, Wireless Communications: Principles and Practice, 1st edition ed.,
Prentice Hall 1996.

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

User-centric smart sensing for non-intrusive electricity consumption disaggregation in buildings

PDF

Radio localization techniques using ranked sequences

PDF

Understanding human-building-emergency interactions in the built environment

PDF

Point cloud data fusion of RGB and thermal information for advanced building envelope modeling in support of energy audits for large districts

PDF

Towards health-conscious spaces: building for human well-being and performance

PDF

Achieving efficient MU-MIMO and indoor localization via switched-beam antennas

PDF

In-situ quality assessment of scan data for as-built models using building-specific geometric features

PDF

Data-driven optimization for indoor localization

PDF

Localization of multiple targets in multi-path environnents

PDF

Distributed algorithms for source localization using quantized sensor readings

PDF

A framework for comprehensive assessment of resilience and other dimensions of asset management in metropolis-scale transport systems

PDF

Semantic modeling of outdoor scenes for the creation of virtual environments and simulations

PDF

Understanding human-building interactions through perceptual decision-making processes

PDF

Multichannel data collection for throughput maximization in wireless sensor networks

PDF

Efficient pipelines for vision-based context sensing

PDF

High-performance distributed computing techniques for wireless IoT and connected vehicle systems

PDF

Application-driven compressed sensing

PDF

Efficient crowd-based visual learning for edge devices

PDF

Differentially private learned models for location services

PDF

Economic model predictive control for building energy systems

Asset Metadata

Creator Li, Nan (author)

Core Title A radio frequency based indoor localization framework for supporting building emergency response operations

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Civil Engineering (Construction Engineering)

Publication Date 04/02/2014

Defense Date 01/21/2014

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag building emergency,emergency response operation,indoor localization,localization algorithm,OAI-PMH Harvest,radio frequency

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Becerik-Gerber, Burcin (committee chair), Krishnamachari, Bhaskar (committee member), Soibelman, Lucio (committee member)

Creator Email nanl@usc.edu,nanliconan@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-372593

Unique identifier UC11297148

Identifier etd-LiNan-2318.pdf (filename),usctheses-c3-372593 (legacy record id)

Legacy Identifier etd-LiNan-2318.pdf

Dmrecord 372593

Document Type Dissertation

Format application/pdf (imt)

Rights Li, Nan

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA