Melika Hozhabri
HUMAN DETECTION AND TRACKING WITH UWB RADAR 2019
Mälardalen University Licentiate Thesis 280
Human Detection and Tracking
with UWB radar
Melika Hozhabri
ISBN 978-91-7485-435-0
ISSN 1651-9256
Address: P.O. Box 883, SE-721 23 Västerås. Sweden
Address: P.O. Box 325, SE-631 05 Eskilstuna. Sweden
E-mail: info@mdh.se Web: www.mdh.se
M¨alardalen University Press Licentiate Thesis
No. 280
HUMAN DETECTION AND TRACKING
WITH UWB RADAR
Melika Hozhabri
2019
School of Innovation, Design and Engineering
M¨alardalen University Press Licentiate Thesis
No. 280
HUMAN DETECTION AND TRACKING
WITH UWB RADAR
Melika Hozhabri
2019
School of Innovation, Design and Engineering
1
Copyright © Melika Hozhabri, 2019
ISSN 1651-9256
ISBN 978-91-7485-435-0
Printed by E-Print AB, Stockholm, Sweden
Copyright © Melika Hozhabri, 2019
ISSN 1651-9256
ISBN 978-91-7485-435-0
Printed by E-Print AB, Stockholm, Sweden
2
To my family To my family
3
4
Abstract
As robots and automated machineries are increasingly replacing the
manual operations, protecting humans who are working in collaboration with these machines is becoming an increasingly important task.
Technologies such as cameras, infra-red and seismic sensors as well as
radar systems are used for presence detection and localization of human
beings. Among different radar sensors, Ultra Wide Band (UWB) radar
has shown some advantages such as providing the distance to the object
with good precision and high performance even under adverse weather
and lightning conditions. In contrary to traditional radar systems which
use a specific frequency and high output power, UWB Radar uses a
wide frequency band (> 500 MHz) and low output power to measure
the distance to the object.
The purpose of this thesis is to investigate UWB radar system for
protecting humans around dangerous machinery in environments like
mines where conditions like dirt, fog, and lack of light cause other technologies such as cameras to have a limited functionality. Experimental
measurements are done to validate the hardware and to investigate its
constraints.
Comparison between two dominant UWB radar technologies is performed: Pulse and M-sequence UWB radar for static human being detection. The results show that M-sequence UWB radar is better suited
for detecting the static human target at larger distances. The better
performance comes at the cost of higher power usage. Measurements of
human walking in different environments is done to measure and compare the background noise and radar reflection of the human body. A
human phantom is developed and choice of material and shape for it is
discussed. The reflection of the phantom is analyzed and compared with
the reflection of a human trunk. Furthermore, the choice of frequency
in discerning human beings is discussed.
Signal processing algorithms and filters are developed for tracking of
the human presence, position and movements. These algorithms contain
pre-processing of the signal such as removing the background, detection
and positioning techniques.
i
Abstract
As robots and automated machineries are increasingly replacing the
manual operations, protecting humans who are working in collaboration with these machines is becoming an increasingly important task.
Technologies such as cameras, infra-red and seismic sensors as well as
radar systems are used for presence detection and localization of human
beings. Among different radar sensors, Ultra Wide Band (UWB) radar
has shown some advantages such as providing the distance to the object
with good precision and high performance even under adverse weather
and lightning conditions. In contrary to traditional radar systems which
use a specific frequency and high output power, UWB Radar uses a
wide frequency band (> 500 MHz) and low output power to measure
the distance to the object.
The purpose of this thesis is to investigate UWB radar system for
protecting humans around dangerous machinery in environments like
mines where conditions like dirt, fog, and lack of light cause other technologies such as cameras to have a limited functionality. Experimental
measurements are done to validate the hardware and to investigate its
constraints.
Comparison between two dominant UWB radar technologies is performed: Pulse and M-sequence UWB radar for static human being detection. The results show that M-sequence UWB radar is better suited
for detecting the static human target at larger distances. The better
performance comes at the cost of higher power usage. Measurements of
human walking in different environments is done to measure and compare the background noise and radar reflection of the human body. A
human phantom is developed and choice of material and shape for it is
discussed. The reflection of the phantom is analyzed and compared with
the reflection of a human trunk. Furthermore, the choice of frequency
in discerning human beings is discussed.
Signal processing algorithms and filters are developed for tracking of
the human presence, position and movements. These algorithms contain
pre-processing of the signal such as removing the background, detection
and positioning techniques.
i
5
6
Sammandrag
I takt med att robotar och automatiska maskiner i ¨okande grad ers¨atter
manuella arbetsuppgifter, ¨okar behovet av skydd f¨or m¨anniskor som arbetar tillsammans med dessa maskiner. Teknologier s˚asom kameror, infrar¨oda och seismiska sensorer samt radarsystem anv¨ands f¨or n¨arvarodetektering och lokalisering av m¨anniskor. Bland olika radarsensorer har
Ultra Wide Band (UWB) radarn visat n˚agra f¨ordelar, s˚asom att ge avst˚and till objektet med god precision och h¨og prestanda ¨aven i ogynnsamma v¨ader- och ljush˚allanden. Till skillnad fr˚an traditionella radarsystem
som anv¨ander en specifik frekvens och h¨og uteffekt, anv¨ander UWB Radar ett brett frekvensband (> 500 MHz) och l˚ag uteffekt f¨or att m¨ata
avst˚and till objekt.
Syftet med den h¨ar avhandlingen ¨ar att anv¨anda UWB-radarsystem
f¨or att skydda m¨anniskor som vistas i n¨arhet av farliga maskiner i milj¨oer
som gruvor, d¨ar f¨orh˚allanden som smuts, dimma och brist p˚a ljus g¨or
att andra tekniker s˚asom kameror f˚ar en minskad funktionalitet. Experimentella m¨atningar g¨ors f¨or att validera h˚ardvaran och f¨or att unders¨oka
dess begr¨ansningar.
J¨amf¨orelse mellan tv˚a dominerande UWB-radarteknologier: Impuls
och M-sekvens UWB-radar f¨or statisk detektering av m¨anniska utf¨ors.
Resultaten visar att M-sekvensen UWB-radar ¨ar b¨attre l¨ampad f¨or att
detektera scenariot med statiska m¨anskliga m˚al p˚a st¨orre avst˚and. B¨attre
prestanda kr¨aver en h¨ogre str¨omf¨orbrukning. M¨atningar av m¨ansk-lig
g˚ang i olika milj¨oer g¨ors f¨or att m¨ata och j¨amf¨ora bakgrundsbrus och
radarreflektion av m¨anniskokroppen. En m¨ansklig modell utvecklas och
materialval och form diskuteras. Reflektionen fr˚an modellen analyseras
och j¨amf¨ors med reflektionen fr˚an en m¨ansklig b˚al. Vidare diskuteras
valet av frekvens f¨or s¨arskiljning av m¨anniskor.
Signalbehandlingsalgoritmer och filter utvecklas f¨or att sp˚ara m¨anniskans n¨arvaro, position och r¨orelser. Dessa algoritmer inneh˚aller f¨orbehandling av signalen s˚asom att eliminering av bakgrunden, detekterings och
positioneringstekniker.
iii
Sammandrag
I takt med att robotar och automatiska maskiner i ¨okande grad ers¨atter
manuella arbetsuppgifter, ¨okar behovet av skydd f¨or m¨anniskor som arbetar tillsammans med dessa maskiner. Teknologier s˚asom kameror, infrar¨oda och seismiska sensorer samt radarsystem anv¨ands f¨or n¨arvarodetektering och lokalisering av m¨anniskor. Bland olika radarsensorer har
Ultra Wide Band (UWB) radarn visat n˚agra f¨ordelar, s˚asom att ge avst˚and till objektet med god precision och h¨og prestanda ¨aven i ogynnsamma v¨ader- och ljush˚allanden. Till skillnad fr˚an traditionella radarsystem
som anv¨ander en specifik frekvens och h¨og uteffekt, anv¨ander UWB Radar ett brett frekvensband (> 500 MHz) och l˚ag uteffekt f¨or att m¨ata
avst˚and till objekt.
Syftet med den h¨ar avhandlingen ¨ar att anv¨anda UWB-radarsystem
f¨or att skydda m¨anniskor som vistas i n¨arhet av farliga maskiner i milj¨oer
som gruvor, d¨ar f¨orh˚allanden som smuts, dimma och brist p˚a ljus g¨or
att andra tekniker s˚asom kameror f˚ar en minskad funktionalitet. Experimentella m¨atningar g¨ors f¨or att validera h˚ardvaran och f¨or att unders¨oka
dess begr¨ansningar.
J¨amf¨orelse mellan tv˚a dominerande UWB-radarteknologier: Impuls
och M-sekvens UWB-radar f¨or statisk detektering av m¨anniska utf¨ors.
Resultaten visar att M-sekvensen UWB-radar ¨ar b¨attre l¨ampad f¨or att
detektera scenariot med statiska m¨anskliga m˚al p˚a st¨orre avst˚and. B¨attre
prestanda kr¨aver en h¨ogre str¨omf¨orbrukning. M¨atningar av m¨ansk-lig
g˚ang i olika milj¨oer g¨ors f¨or att m¨ata och j¨amf¨ora bakgrundsbrus och
radarreflektion av m¨anniskokroppen. En m¨ansklig modell utvecklas och
materialval och form diskuteras. Reflektionen fr˚an modellen analyseras
och j¨amf¨ors med reflektionen fr˚an en m¨ansklig b˚al. Vidare diskuteras
valet av frekvens f¨or s¨arskiljning av m¨anniskor.
Signalbehandlingsalgoritmer och filter utvecklas f¨or att sp˚ara m¨anniskans n¨arvaro, position och r¨orelser. Dessa algoritmer inneh˚aller f¨orbehandling av signalen s˚asom att eliminering av bakgrunden, detekterings och
positioneringstekniker.
iii
7
8
Acknowledgements
Finally I have reached this half way milestone. It was a stimulating, educating and exciting journey alongside feelings of frustration, confusion
and failures. But nothing worth having comes easy.
I would like to thank:
Bj¨orn, my wonderful and supportive partner. Putting up with me on
stressful days, nights, holidays and weekends with your amazing patience
and understanding. It would have not been possible without you.
My parents for supporting me through my education and believing
in me. Giving me all they could to support my interest in science and
technology.
My former manager Dag Lindahl that started this project with a
great vision and positiveness.
My supervisors, Maria Lind`en, Nikola Petrovi´c and Martin Ekstr¨om
for their help, support and comments during this thesis.
Per Olov Risman for his engagement and comments during this thesis.
I appreciate your knowledge, engagement and being available at any
time.
My colleagues and friends at MDH, Arash Gharebaghi for your support, help and comments, and Zeinab Bakhshi for your positive spirit.
ITS-EASY team Kristina Lundkvist, Radu Dobrin, Gunnar Widforss, and Malin Rosqvist for the great discussions, trips and support.
Addiva AB, Bj¨orn Lindstr¨om and my manager Rasmus Fridberg for
providing me this opportunity to learn and grow.
My colleagues at Addiva AB, Nils Brynedal Ignell, Olle Wedin, Ralf
Str¨omberg, Iryna Gusstavsson, and Simon Johansson for the great team
work, help, discussions and ideas.
My former colleagues at RISE SICS, Ali Balador who I have learned
a lot from during my research in SafePOS project, and Markus Bohlin
for the great inspiration you gave me during a short meeting.
All of my friends, the valuable treasure I have. For letting me know
that I have your help and support. Especial thanks to Ella for listening,
understanding and all the encouragements on daily basis.
At last, my son who shines like a sun and greets me with laughter
and hugs every day. You gives me a reason to carry on.
Melika Hozhabri, V¨aster˚as, September, 2019
Acknowledgements
Finally I have reached this half way milestone. It was a stimulating, educating and exciting journey alongside feelings of frustration, confusion
and failures. But nothing worth having comes easy.
I would like to thank:
Bj¨orn, my wonderful and supportive partner. Putting up with me on
stressful days, nights, holidays and weekends with your amazing patience
and understanding. It would have not been possible without you.
My parents for supporting me through my education and believing
in me. Giving me all they could to support my interest in science and
technology.
My former manager Dag Lindahl that started this project with a
great vision and positiveness.
My supervisors, Maria Lind`en, Nikola Petrovi´c and Martin Ekstr¨om
for their help, support and comments during this thesis.
Per Olov Risman for his engagement and comments during this thesis.
I appreciate your knowledge, engagement and being available at any
time.
My colleagues and friends at MDH, Arash Gharebaghi for your support, help and comments, and Zeinab Bakhshi for your positive spirit.
ITS-EASY team Kristina Lundkvist, Radu Dobrin, Gunnar Widforss, and Malin Rosqvist for the great discussions, trips and support.
Addiva AB, Bj¨orn Lindstr¨om and my manager Rasmus Fridberg for
providing me this opportunity to learn and grow.
My colleagues at Addiva AB, Nils Brynedal Ignell, Olle Wedin, Ralf
Str¨omberg, Iryna Gusstavsson, and Simon Johansson for the great team
work, help, discussions and ideas.
My former colleagues at RISE SICS, Ali Balador who I have learned
a lot from during my research in SafePOS project, and Markus Bohlin
for the great inspiration you gave me during a short meeting.
All of my friends, the valuable treasure I have. For letting me know
that I have your help and support. Especial thanks to Ella for listening,
understanding and all the encouragements on daily basis.
At last, my son who shines like a sun and greets me with laughter
and hugs every day. You gives me a reason to carry on.
Melika Hozhabri, V¨aster˚as, September, 2019
9
10
Contents
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . 2
1.3 Research Method . . . . . . . . . . . . . . . . . . . . . . . 3
2 Related Work 5
2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 UWB Radar System . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Hardware Platform . . . . . . . . . . . . . . . . . . 9
2.2.2 Ultra Wide-band Antennas . . . . . . . . . . . . . 9
2.3 Signal Processing . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Clutter Removal . . . . . . . . . . . . . . . . . . . 11
2.3.2 Detection . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.3 Localization . . . . . . . . . . . . . . . . . . . . . . 13
2.3.4 Tracking . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 Human Radar Cross Section . . . . . . . . . . . . . . . . . 14
2.5 Related Experimental Systems . . . . . . . . . . . . . . . 15
2.5.1 Systems for Human Detection in LOS . . . . . . . 15
2.5.2 Systems for Human Detection Behind Obstacles . 18
3 System Design and Validation 22
3.1 Software Platform Design . . . . . . . . . . . . . . . . . . 22
3.2 System Validation and Measurements . . . . . . . . . . . 23
3.2.1 The Vivaldi Antenna Radiation Pattern Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.2 Comparison of M-sequence vs Pulse radar for Static
Human Detection (Paper A) . . . . . . . . . . . . 27
vii
Contents
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . 2
1.3 Research Method . . . . . . . . . . . . . . . . . . . . . . . 3
2 Related Work 5
2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 UWB Radar System . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Hardware Platform . . . . . . . . . . . . . . . . . . 9
2.2.2 Ultra Wide-band Antennas . . . . . . . . . . . . . 9
2.3 Signal Processing . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Clutter Removal . . . . . . . . . . . . . . . . . . . 11
2.3.2 Detection . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.3 Localization . . . . . . . . . . . . . . . . . . . . . . 13
2.3.4 Tracking . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 Human Radar Cross Section . . . . . . . . . . . . . . . . . 14
2.5 Related Experimental Systems . . . . . . . . . . . . . . . 15
2.5.1 Systems for Human Detection in LOS . . . . . . . 15
2.5.2 Systems for Human Detection Behind Obstacles . 18
3 System Design and Validation 22
3.1 Software Platform Design . . . . . . . . . . . . . . . . . . 22
3.2 System Validation and Measurements . . . . . . . . . . . 23
3.2.1 The Vivaldi Antenna Radiation Pattern Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.2 Comparison of M-sequence vs Pulse radar for Static
Human Detection (Paper A) . . . . . . . . . . . . 27
vii
11
viii Contents
3.2.3 Walking Human Detection in Different Environments (Paper B) . . . . . . . . . . . . . . . . . . . 27
3.2.4 Respiration Simulation and Measurement . . . . . 28
3.2.5 Phantom Measurements (Paper C) . . . . . . . . . 29
3.3 Signal Processing Algorithms . . . . . . . . . . . . . . . . 31
3.3.1 Pre-processing . . . . . . . . . . . . . . . . . . . . 31
3.3.2 Clutter Removal . . . . . . . . . . . . . . . . . . . 32
3.3.3 Target detection . . . . . . . . . . . . . . . . . . . 34
3.3.4 Target Tracking . . . . . . . . . . . . . . . . . . . 35
3.3.5 Other Algorithms . . . . . . . . . . . . . . . . . . . 36
4 Contribution 38
5 Conclusion and Future Work 41
5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . 42
A Abbrevations 44
Bibliography 47
viii Contents
3.2.3 Walking Human Detection in Different Environments (Paper B) . . . . . . . . . . . . . . . . . . . 27
3.2.4 Respiration Simulation and Measurement . . . . . 28
3.2.5 Phantom Measurements (Paper C) . . . . . . . . . 29
3.3 Signal Processing Algorithms . . . . . . . . . . . . . . . . 31
3.3.1 Pre-processing . . . . . . . . . . . . . . . . . . . . 31
3.3.2 Clutter Removal . . . . . . . . . . . . . . . . . . . 32
3.3.3 Target detection . . . . . . . . . . . . . . . . . . . 34
3.3.4 Target Tracking . . . . . . . . . . . . . . . . . . . 35
3.3.5 Other Algorithms . . . . . . . . . . . . . . . . . . . 36
4 Contribution 38
5 Conclusion and Future Work 41
5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . 42
A Abbrevations 44
Bibliography 47
12
Chapter 1
Introduction
Human-machine interaction is becoming more important because of its
potential military, safety, security, and entertainment applications. The
manual operating machines are getting replaced with robots and automated machineries and humans needs to be able to safely work in collaboration with them. Furthermore, the emerging market for autonomous
vehicles demands reliable pedestrian detection and localization. Meanwhile recent advances in technology makes it possible to detect, localize
and respond to the presence of a human. Techniques such as Global
Positioning System (GPS), infrared detectors, vision based systems, vibration and seismic sensors, acoustics sensors, and radar systems are
used to achieve human detection and localization.
Among different radar sensors, Ultra Wide Band (UWB) radar offers
high-resolution ranging in dynamic environments, low power consumption and high performance in multipath channel without requiring Line
Of Sight (LOS).
Several efforts by different research groups have been performed within
the domain of detecting human targets with UWB radar [1–4]. UWB
radar has the capability to penetrate most common building materials,
therefore many researchers aim at detection and tracking of humans behind walls for surveillance and rescue operations [5–9].
In this thesis, UWB radar is used for detection, localization, and
tracking of humans to provide reliable pedestrian safety in the presence
of moving machines. The UWB radar system constraints and capabilities
are examined by performing measurements in different environments and
1
Chapter 1
Introduction
Human-machine interaction is becoming more important because of its
potential military, safety, security, and entertainment applications. The
manual operating machines are getting replaced with robots and automated machineries and humans needs to be able to safely work in collaboration with them. Furthermore, the emerging market for autonomous
vehicles demands reliable pedestrian detection and localization. Meanwhile recent advances in technology makes it possible to detect, localize
and respond to the presence of a human. Techniques such as Global
Positioning System (GPS), infrared detectors, vision based systems, vibration and seismic sensors, acoustics sensors, and radar systems are
used to achieve human detection and localization.
Among different radar sensors, Ultra Wide Band (UWB) radar offers
high-resolution ranging in dynamic environments, low power consumption and high performance in multipath channel without requiring Line
Of Sight (LOS).
Several efforts by different research groups have been performed within
the domain of detecting human targets with UWB radar [1–4]. UWB
radar has the capability to penetrate most common building materials,
therefore many researchers aim at detection and tracking of humans behind walls for surveillance and rescue operations [5–9].
In this thesis, UWB radar is used for detection, localization, and
tracking of humans to provide reliable pedestrian safety in the presence
of moving machines. The UWB radar system constraints and capabilities
are examined by performing measurements in different environments and
1
13
2 Chapter 1. Introduction
set-ups. The measurements were later processed by signal processing
algorithms to extract the human target signal.
1.1 Motivation
Safety is an important consideration in human-machine interactions. Industrial machines can move or have moving parts that can cause hazards
to humans surrounding them. Hazardous industrial machines and operators are sometimes separated by barriers to avoid any contact between
them and as a consequence the productivity of the site is reduced [10].
Most state of the art machineries and robots are equipped with sensors
for collision avoidance such as laser scanners, cameras, and infrared sensors [11]. These systems are used for detection of objects and obstacles,
their position, and their relative speed to the machine or robot [10].
They are performing reliably in some conditions but suffer from a number of limitations because of the optical technology they rely on. Some
factors such as large open areas, fog, smoke, dust, dirt, condensation,
lighting conditions, and reflections may cause faulty sensor values. In
addition, visual, infrared, and seismic sensors need to be placed in close
proximity to the target whereas radar sensors depend on the frequency
of the operation and can function up to several hundred meters.
1.2 Problem Formulation
The purpose of this research is to develop a system for protecting humans around hazardous machinery in environments such as mines where
conditions such as dirt, fog and lack of light cause problems with other
technologies. This feature is needed in industries where safety requirements around automatic machineries are getting more stringent. The
research goal for this thesis is:
”To evaluate a wireless system and to develop signal processing algorithms able to detect and localize humans in enclosed environments.”
This general goal can be split up into the following research questions.
RQ1 What are the UWB radar system characteristics and constraints in human detection applications?
The system needs to be carefully analysed with regard to its throughput and its limitations. Signal and noise characteristics need to
2 Chapter 1. Introduction
set-ups. The measurements were later processed by signal processing
algorithms to extract the human target signal.
1.1 Motivation
Safety is an important consideration in human-machine interactions. Industrial machines can move or have moving parts that can cause hazards
to humans surrounding them. Hazardous industrial machines and operators are sometimes separated by barriers to avoid any contact between
them and as a consequence the productivity of the site is reduced [10].
Most state of the art machineries and robots are equipped with sensors
for collision avoidance such as laser scanners, cameras, and infrared sensors [11]. These systems are used for detection of objects and obstacles,
their position, and their relative speed to the machine or robot [10].
They are performing reliably in some conditions but suffer from a number of limitations because of the optical technology they rely on. Some
factors such as large open areas, fog, smoke, dust, dirt, condensation,
lighting conditions, and reflections may cause faulty sensor values. In
addition, visual, infrared, and seismic sensors need to be placed in close
proximity to the target whereas radar sensors depend on the frequency
of the operation and can function up to several hundred meters.
1.2 Problem Formulation
The purpose of this research is to develop a system for protecting humans around hazardous machinery in environments such as mines where
conditions such as dirt, fog and lack of light cause problems with other
technologies. This feature is needed in industries where safety requirements around automatic machineries are getting more stringent. The
research goal for this thesis is:
”To evaluate a wireless system and to develop signal processing algorithms able to detect and localize humans in enclosed environments.”
This general goal can be split up into the following research questions.
RQ1 What are the UWB radar system characteristics and constraints in human detection applications?
The system needs to be carefully analysed with regard to its throughput and its limitations. Signal and noise characteristics need to
14
1.3 Research Method 3
be obtained by carefully planned measurements in different environments. Furthermore, understanding the characteristics of the
background noise will help to know how much it will affect the
signal detection.
RQ2 What are the most appropriate signal processing algorithms to be used for detection of humans, using the chosen sensor system?
To be able to extract the human target signal, raw radar data
passes several signal-processing steps. This signal usually is affected by noise, clutter and attenuation. Signal processing algorithms and their order affect the processing power and computational complexity. There are some decisions to be made such
as what and how many features that should be extracted. The
algorithms should also be examined for false alarms and missed
detection.
RQ3 Is it possible to make a phantom of a human in order to
obtain controlled conditions in measurements?
In measurement of real humans, the results may vary based on
the body size, orientation, posture, and clothing. Performing measurements with a phantom of human body reduces the unwanted
variabilities of real human measurements. Which material combinations and geometry can mimic a radar cross section of a human?
1.3 Research Method
This work started by a thorough literature review in order to gain knowledge about the state of the art and to recognize the requirements and
challenges. In addition, collaborations between Addiva AB and industrial partners in robotics and mining industry helped us to achieve a
better understanding of the real world challenges that each particular
industry is facing. One prerequisite of the work was to use an already
existing hardware system for the UWB radar. A part of this research is
about understanding the hardware, its advantages and limitations.
The research method used in this thesis is based on the experimental
results lead to validating the theory. Series of experiments are performed
and collected data are analyzed. This resulted in scientific papers and
1.3 Research Method 3
be obtained by carefully planned measurements in different environments. Furthermore, understanding the characteristics of the
background noise will help to know how much it will affect the
signal detection.
RQ2 What are the most appropriate signal processing algorithms to be used for detection of humans, using the chosen sensor system?
To be able to extract the human target signal, raw radar data
passes several signal-processing steps. This signal usually is affected by noise, clutter and attenuation. Signal processing algorithms and their order affect the processing power and computational complexity. There are some decisions to be made such
as what and how many features that should be extracted. The
algorithms should also be examined for false alarms and missed
detection.
RQ3 Is it possible to make a phantom of a human in order to
obtain controlled conditions in measurements?
In measurement of real humans, the results may vary based on
the body size, orientation, posture, and clothing. Performing measurements with a phantom of human body reduces the unwanted
variabilities of real human measurements. Which material combinations and geometry can mimic a radar cross section of a human?
1.3 Research Method
This work started by a thorough literature review in order to gain knowledge about the state of the art and to recognize the requirements and
challenges. In addition, collaborations between Addiva AB and industrial partners in robotics and mining industry helped us to achieve a
better understanding of the real world challenges that each particular
industry is facing. One prerequisite of the work was to use an already
existing hardware system for the UWB radar. A part of this research is
about understanding the hardware, its advantages and limitations.
The research method used in this thesis is based on the experimental
results lead to validating the theory. Series of experiments are performed
and collected data are analyzed. This resulted in scientific papers and
15
4 Chapter 1. Introduction
better understanding of the system. This has been an iterative process
as each result led to more knowledge which in turn resulted in new
literature studies and set of experiments.1
During the experimental research process, the system characteristics
is thoroughly investigated and measured. This includes investigating the
ability of the sensor to detect static and dynamic humans and to measure
the effect of the environment noise on the signal detection. Thereafter,
suitable signal processing algorithms are developed and implemented.
1This research is partly financed by Vinnova: Diarienummer 2014-00484
4 Chapter 1. Introduction
better understanding of the system. This has been an iterative process
as each result led to more knowledge which in turn resulted in new
literature studies and set of experiments.1
During the experimental research process, the system characteristics
is thoroughly investigated and measured. This includes investigating the
ability of the sensor to detect static and dynamic humans and to measure
the effect of the environment noise on the signal detection. Thereafter,
suitable signal processing algorithms are developed and implemented.
1This research is partly financed by Vinnova: Diarienummer 2014-00484
16
Chapter 2
Related Work
This chapter presents an overview of related research in the area of human detection and tracking with UWB radar. Firstly, a background of
human detection and tracking with UWB radar is presented. Thereafter
follows a short introduction of UWB radar systems, signal processing
steps, and a brief explanation of human radar cross section. Finally,
experimental systems using UWB radar for human presence detection
and localization are presented. Table 2.1 summarizes these systems, including their properties as accuracy, hardware, measurement set-up and
signal processing methods.
2.1 Background
Human detection and tracking consists of measuring spatio-temporal
properties of humans such as presence, count, location, track, and identity [12]. Measurable human traits with UWB radar are shown in Figure
2.1. By using UWB radar, the Radar Cross Section (RCS) (see section
2.4) of a human body and external body-part motions such as walking
and hands and feet movements can be detected. Also internal body-part
motions, for example involuntary motion of internal organs such as heart
beat and respiration, can be measured.
UWB radar was used for human detection in LOS [3, 13, 14] as well
as detection of humans behind obstacles and walls due to its capability
to penetrate most common building materials [4, 6–8]. Both methods
require signal processing algorithms and techniques for human detection,
5
Chapter 2
Related Work
This chapter presents an overview of related research in the area of human detection and tracking with UWB radar. Firstly, a background of
human detection and tracking with UWB radar is presented. Thereafter
follows a short introduction of UWB radar systems, signal processing
steps, and a brief explanation of human radar cross section. Finally,
experimental systems using UWB radar for human presence detection
and localization are presented. Table 2.1 summarizes these systems, including their properties as accuracy, hardware, measurement set-up and
signal processing methods.
2.1 Background
Human detection and tracking consists of measuring spatio-temporal
properties of humans such as presence, count, location, track, and identity [12]. Measurable human traits with UWB radar are shown in Figure
2.1. By using UWB radar, the Radar Cross Section (RCS) (see section
2.4) of a human body and external body-part motions such as walking
and hands and feet movements can be detected. Also internal body-part
motions, for example involuntary motion of internal organs such as heart
beat and respiration, can be measured.
UWB radar was used for human detection in LOS [3, 13, 14] as well
as detection of humans behind obstacles and walls due to its capability
to penetrate most common building materials [4, 6–8]. Both methods
require signal processing algorithms and techniques for human detection,
5
17
6 Chapter 2. Related Work
Figure 2.1: Measurable human traits with UWB radar.
and the through wall detection requires additional algorithms for clutter
removal due to the existence of the obstacle or wall. The wall distorts
the radar signature in form of attenuation, refraction, and multipath,
therefore the wall effect must be compensated for by using the known
parameters such as thickness and relative permittivity of the wall. This
thesis focuses on detecting human beings in LOS with UWB Radar in
cluttered environments.
2.2 UWB Radar System
Based on the definition by the US Federal Communications Commission’s (FCC), a signal can be defined as an UWB signal if its bandwidth
is greater than 500 MHz or its fractional bandwidth is greater than 0.2
where fractional bandwidth is defined in equation 2.1.
Bf = 2
fH − fL
fH + fL
(2.1)
Different types of UWB radar are developed and based on the excitation wave-form they are divided into different categories. Three of the
most prominent categories are Frequency Modulated Continuous Wave
(FMCW) UWB radar, pulse UWB radar, and M-sequence UWB radar.
6 Chapter 2. Related Work
Figure 2.1: Measurable human traits with UWB radar.
and the through wall detection requires additional algorithms for clutter
removal due to the existence of the obstacle or wall. The wall distorts
the radar signature in form of attenuation, refraction, and multipath,
therefore the wall effect must be compensated for by using the known
parameters such as thickness and relative permittivity of the wall. This
thesis focuses on detecting human beings in LOS with UWB Radar in
cluttered environments.
2.2 UWB Radar System
Based on the definition by the US Federal Communications Commission’s (FCC), a signal can be defined as an UWB signal if its bandwidth
is greater than 500 MHz or its fractional bandwidth is greater than 0.2
where fractional bandwidth is defined in equation 2.1.
Bf = 2
fH − fL
fH + fL
(2.1)
Different types of UWB radar are developed and based on the excitation wave-form they are divided into different categories. Three of the
most prominent categories are Frequency Modulated Continuous Wave
(FMCW) UWB radar, pulse UWB radar, and M-sequence UWB radar.
18
2.2 UWB Radar System 7
Figure 2.2: A basic scheme of an M-sequence radar [15].
In this thesis, we have used a commercially available M-sequence UWB
radar which has been developed by Radarbolaget (G¨avle, Sweden). The
company is primarily working with radar systems for real time, and
through-the-wall monitoring of furnaces1
.
A basic scheme of a M-sequence UWB radar is shown in Figure 2.2.
This system were chosen for its flexibility and scalability. The radar
system is coded on a FPGA, therefore it is possible to change its design
on demand, which makes it flexible. Another advantage is the scalability
which allows connection of up to twelve antenna pair sensors to the
system.
Data Representation
The radar system is able to record hundreds of scans per second. Theses
recorded scans can be presented as radar return at a certain time as it
is shown in Figure 2.3. The horizontal axis represents the time of flight,
which is twice the distance to the object in detection region of the radar.
The vertical axis represents the correlation of received and transmitted
signal in the RPU. The distance between two adjacent measurement
points in horizontal axis is 22367.7 ns corresponds to 4166.7 µm. The
radar scans can be stacked along the time axis. This results in a twodimensional graph, which is called a radargram. The structure of the
radargram is shown in Figure 2.4.
1http://www.radarbolaget.com
2.2 UWB Radar System 7
Figure 2.2: A basic scheme of an M-sequence radar [15].
In this thesis, we have used a commercially available M-sequence UWB
radar which has been developed by Radarbolaget (G¨avle, Sweden). The
company is primarily working with radar systems for real time, and
through-the-wall monitoring of furnaces1
.
A basic scheme of a M-sequence UWB radar is shown in Figure 2.2.
This system were chosen for its flexibility and scalability. The radar
system is coded on a FPGA, therefore it is possible to change its design
on demand, which makes it flexible. Another advantage is the scalability
which allows connection of up to twelve antenna pair sensors to the
system.
Data Representation
The radar system is able to record hundreds of scans per second. Theses
recorded scans can be presented as radar return at a certain time as it
is shown in Figure 2.3. The horizontal axis represents the time of flight,
which is twice the distance to the object in detection region of the radar.
The vertical axis represents the correlation of received and transmitted
signal in the RPU. The distance between two adjacent measurement
points in horizontal axis is 22367.7 ns corresponds to 4166.7 µm. The
radar scans can be stacked along the time axis. This results in a twodimensional graph, which is called a radargram. The structure of the
radargram is shown in Figure 2.4.
1http://www.radarbolaget.com
19
8 Chapter 2. Related Work
Figure 2.3: Radar return signal. The first peak represents the antenna
cross talk (mutual coupling) and the second peak is an object reflection
which is in detection region of the radar
Figure 2.4: Radargram structure: Each raw (T11 to T1m) represents the
radar return signal in a fast time. The radar return signals are stacked
along n rows in slow time. ∆d represents the distance between two
consecutive samples of radar return that specifies the range resolution.
8 Chapter 2. Related Work
Figure 2.3: Radar return signal. The first peak represents the antenna
cross talk (mutual coupling) and the second peak is an object reflection
which is in detection region of the radar
Figure 2.4: Radargram structure: Each raw (T11 to T1m) represents the
radar return signal in a fast time. The radar return signals are stacked
along n rows in slow time. ∆d represents the distance between two
consecutive samples of radar return that specifies the range resolution.
20
2.2 UWB Radar System 9
2.2.1 Hardware Platform
The M-sequence UWB radar from Radarbolaget is shown in figure 2.5.
The hardware platform consists of a Radar Processing Unit (RPU), a
Wide-band Radar Transceiver (WRT) and a pair of Vivaldi antennas.
The RPU is responsible for processing and synchronization of radar signals. The WRT is a device that converts the radar signals and directs
them to the RPU. One of the Vivaldi antennas are acting as a transmitter, sending the M-sequence code, while the other antenna is receiving
the reflected signal and at the same time the signal is AD-converted by
an AD-converter in the antenna. There is also a data switch that keeps
track of and addresses each antenna. To each WRT, two to twelve antenna pairs can be coupled. The transmit gain is adjustable and has a
maximum value of −10 dBm and the radar bandwidth is approximately
2 GHz (1–3 GHz). For more detailed hardware description see Radarbolaget’s website 2 and paper A. The WRT and RPU are placed in a
computer case for easier supply of power and handling. The WRT and
RPU are connected to each other with an optical fiber. The RPU is
connected with an USB connection to the PC. The antenna casing is
entirely plastic and manufactured by a 3D printer.
2.2.2 Ultra Wide-band Antennas
An UWB radar system requires an antenna capable of receiving a wide
frequency range at the same time. Thus, the antenna behavior and
performance must be consistent and predictable across the entire bandwidth. Ideally, pattern and matching should be stable across the entire
band width and the antenna should preferably has a fixed phase center [16]. Furthermore, the frequency-dependent characteristics of the antennas and the time-domain effects and properties have to be known [17].
There are different types of antennas used in UWB systems such
as traveling-wave structures, frequency-independent antennas, multiple
resonance antennas, and electrically small antennas.
Vivaldi Antenna
There exists several types of UWB antennas for different UWB applications. In telecommunication applications, omni-directional antennas are
2https://www.radarbolaget.com/
2.2 UWB Radar System 9
2.2.1 Hardware Platform
The M-sequence UWB radar from Radarbolaget is shown in figure 2.5.
The hardware platform consists of a Radar Processing Unit (RPU), a
Wide-band Radar Transceiver (WRT) and a pair of Vivaldi antennas.
The RPU is responsible for processing and synchronization of radar signals. The WRT is a device that converts the radar signals and directs
them to the RPU. One of the Vivaldi antennas are acting as a transmitter, sending the M-sequence code, while the other antenna is receiving
the reflected signal and at the same time the signal is AD-converted by
an AD-converter in the antenna. There is also a data switch that keeps
track of and addresses each antenna. To each WRT, two to twelve antenna pairs can be coupled. The transmit gain is adjustable and has a
maximum value of −10 dBm and the radar bandwidth is approximately
2 GHz (1–3 GHz). For more detailed hardware description see Radarbolaget’s website 2 and paper A. The WRT and RPU are placed in a
computer case for easier supply of power and handling. The WRT and
RPU are connected to each other with an optical fiber. The RPU is
connected with an USB connection to the PC. The antenna casing is
entirely plastic and manufactured by a 3D printer.
2.2.2 Ultra Wide-band Antennas
An UWB radar system requires an antenna capable of receiving a wide
frequency range at the same time. Thus, the antenna behavior and
performance must be consistent and predictable across the entire bandwidth. Ideally, pattern and matching should be stable across the entire
band width and the antenna should preferably has a fixed phase center [16]. Furthermore, the frequency-dependent characteristics of the antennas and the time-domain effects and properties have to be known [17].
There are different types of antennas used in UWB systems such
as traveling-wave structures, frequency-independent antennas, multiple
resonance antennas, and electrically small antennas.
Vivaldi Antenna
There exists several types of UWB antennas for different UWB applications. In telecommunication applications, omni-directional antennas are
2https://www.radarbolaget.com/
21
10 Chapter 2. Related Work
Figure 2.5: UWB radar measurement system: The WRT and RPU in
a PC housing.
preferred, but in detection and positioning applications, directive UWB
antennas are used. The antenna in the current system is an antipodal
Vivaldi antenna from Radarbolaget, which has been developed by empirical testing, and therefore the design is not parameterized [18]. The
Vivaldi antenna has advantages over other types of UWB antennas, such
as horn antenna and log periodic array, due to its high directivity, large
bandwidth, small group delay and low cost of manufacturing (it is easy
to fabricate the antenna on a circuit board).
In this thesis a pair of Vivaldi antennas developed in the University
of G¨avle are used in the radar system. The Vivaldi antenna guides the
wave from the feed in a slot line to the exponential wide-band taper which
radiates all the frequencies of the entire bandwidth. The taper shape can
be designed so it provides a smooth transition of the guided wave into the
free space. The Vivaldi antenna has relatively low distortion compared
to other UWB antennas.
10 Chapter 2. Related Work
Figure 2.5: UWB radar measurement system: The WRT and RPU in
a PC housing.
preferred, but in detection and positioning applications, directive UWB
antennas are used. The antenna in the current system is an antipodal
Vivaldi antenna from Radarbolaget, which has been developed by empirical testing, and therefore the design is not parameterized [18]. The
Vivaldi antenna has advantages over other types of UWB antennas, such
as horn antenna and log periodic array, due to its high directivity, large
bandwidth, small group delay and low cost of manufacturing (it is easy
to fabricate the antenna on a circuit board).
In this thesis a pair of Vivaldi antennas developed in the University
of G¨avle are used in the radar system. The Vivaldi antenna guides the
wave from the feed in a slot line to the exponential wide-band taper which
radiates all the frequencies of the entire bandwidth. The taper shape can
be designed so it provides a smooth transition of the guided wave into the
free space. The Vivaldi antenna has relatively low distortion compared
to other UWB antennas.
22
2.3 Signal Processing 11
Figure 2.6: UWB Radar Signal Processing Steps.
2.3 Signal Processing
Rovaˇnkov´a and Kocur [19] presented UWB radar signal processing steps
for human detection and tracking. The different signal processing steps
are presented in Figure 2.6. The signal processing algorithms that are
implemented and used in this thesis are presented in chapter 3.
2.3.1 Clutter Removal
Clutter is an unwanted radar return signals. The goal of this thesis is
to develop a system for human detection which eventually will be used
in highly cluttered environments such as mines, which requires that the
effect of clutter on the detected signal to be investigated. Clutter contains the antenna cross-talk (A direct wave propagating from the transmitter antenna Tx directly to the receiver antenna Rx) and reflections
from other objects than the human in the scenario, such as walls and
machineries. Sources of clutter can also be out-of-band interfering signals at frequencies other than dedicated bandwidth for the system or
in-band interference and thermal noise. Out-of-band interference can be
detected and removed by traditional techniques such as Fourier transformation and band-pass filtering. In-band interference and thermal noise
are harder to identify and remove. Clutter can affect the probability
of detection and the accuracy. To understand and quantify clutter, the
statistical properties of the clutter are often used. A statistical radar
clutter model for modern high resolution radars is presented in [20].
Densities such as Weibull or log-normal distributions were shown to provide reasonable fits for measured clutter densities. In equation [2.2] the
Probability Density Function (PDF) of log normal distribution is shown.
fX(x; µ, σ) = 1
xσ√
2π
e
−
(ln x−µ)
2
2σ2 , x > 0 (2.2)
2.3 Signal Processing 11
Figure 2.6: UWB Radar Signal Processing Steps.
2.3 Signal Processing
Rovaˇnkov´a and Kocur [19] presented UWB radar signal processing steps
for human detection and tracking. The different signal processing steps
are presented in Figure 2.6. The signal processing algorithms that are
implemented and used in this thesis are presented in chapter 3.
2.3.1 Clutter Removal
Clutter is an unwanted radar return signals. The goal of this thesis is
to develop a system for human detection which eventually will be used
in highly cluttered environments such as mines, which requires that the
effect of clutter on the detected signal to be investigated. Clutter contains the antenna cross-talk (A direct wave propagating from the transmitter antenna Tx directly to the receiver antenna Rx) and reflections
from other objects than the human in the scenario, such as walls and
machineries. Sources of clutter can also be out-of-band interfering signals at frequencies other than dedicated bandwidth for the system or
in-band interference and thermal noise. Out-of-band interference can be
detected and removed by traditional techniques such as Fourier transformation and band-pass filtering. In-band interference and thermal noise
are harder to identify and remove. Clutter can affect the probability
of detection and the accuracy. To understand and quantify clutter, the
statistical properties of the clutter are often used. A statistical radar
clutter model for modern high resolution radars is presented in [20].
Densities such as Weibull or log-normal distributions were shown to provide reasonable fits for measured clutter densities. In equation [2.2] the
Probability Density Function (PDF) of log normal distribution is shown.
fX(x; µ, σ) = 1
xσ√
2π
e
−
(ln x−µ)
2
2σ2 , x > 0 (2.2)
23
12 Chapter 2. Related Work
where µ and σ are the mean and standard deviation of the variable x
natural logarithm respectively. The variable x is the value of the clutter
at every point in radar return echo signal. In this thesis, the result of
log-normal fit of the clutter distribution in a semi-anechoic chamber and
the office is presented and compared.
Clutter removal techniques have different complexity based on application, speed, accuracy, and memory requirement. One common method
concerning detection of static objects is background removal. In this
method a radar measurement of the background, before the object of interest is placed in radar detection region, is removed from the raw radar
data after the object is placed in the radar detection region [21]. This
method has a drawback since it does not consider the shadowing phenomena. Shadowing means that the existence of the object will change
the reflection from all the other objects in the scenario.
Exponential averaging is another clutter removal method used with
moving objects. This has advantages such as low complexity and good
performance [6]. In this thesis an exponential algorithm is implemented
and used for detection of dynamic humans. This algorithm is described
in more detail in section 3.3.2.
2.3.2 Detection
In the detection step a decision is being made by the detection algorithm
whether if the signal scattered from the target is absent or present in
the radar data.
Based on Figure 2.1 the human can be detected by UWB radar either
by the motion or RCS. A moving person causes a frequency shift in the
radar echo signal due to Doppler effect. However, humans have other
vibrations and rotations such as swing of the arms while walking. These
micro scale movements produce additional Doppler shifts, referred to
as micro-Doppler effect. Micro-Doppler processing can detect periodic
motions such as movement of arms and legs or respiration [22].
UWB radar provides high-resolution range profile as well as high resolution Doppler spectra therefore it is a good candidate for measurement
of periodic movements of the human body. If there are other moving
objects in detection region of the radar the detection will be more demanding and there will be a need for object recognition afterwards.
V.Chen [23] discussed the radar back-scattering from a walking human model. Based on the model a motion trajectory and velocity pattern
12 Chapter 2. Related Work
where µ and σ are the mean and standard deviation of the variable x
natural logarithm respectively. The variable x is the value of the clutter
at every point in radar return echo signal. In this thesis, the result of
log-normal fit of the clutter distribution in a semi-anechoic chamber and
the office is presented and compared.
Clutter removal techniques have different complexity based on application, speed, accuracy, and memory requirement. One common method
concerning detection of static objects is background removal. In this
method a radar measurement of the background, before the object of interest is placed in radar detection region, is removed from the raw radar
data after the object is placed in the radar detection region [21]. This
method has a drawback since it does not consider the shadowing phenomena. Shadowing means that the existence of the object will change
the reflection from all the other objects in the scenario.
Exponential averaging is another clutter removal method used with
moving objects. This has advantages such as low complexity and good
performance [6]. In this thesis an exponential algorithm is implemented
and used for detection of dynamic humans. This algorithm is described
in more detail in section 3.3.2.
2.3.2 Detection
In the detection step a decision is being made by the detection algorithm
whether if the signal scattered from the target is absent or present in
the radar data.
Based on Figure 2.1 the human can be detected by UWB radar either
by the motion or RCS. A moving person causes a frequency shift in the
radar echo signal due to Doppler effect. However, humans have other
vibrations and rotations such as swing of the arms while walking. These
micro scale movements produce additional Doppler shifts, referred to
as micro-Doppler effect. Micro-Doppler processing can detect periodic
motions such as movement of arms and legs or respiration [22].
UWB radar provides high-resolution range profile as well as high resolution Doppler spectra therefore it is a good candidate for measurement
of periodic movements of the human body. If there are other moving
objects in detection region of the radar the detection will be more demanding and there will be a need for object recognition afterwards.
V.Chen [23] discussed the radar back-scattering from a walking human model. Based on the model a motion trajectory and velocity pattern
24
2.3 Signal Processing 13
of the human body parts was derived. The time-frequency transform of
the signal was then analyzed to derive the micro-Doppler signature of human movement and complex arm and motion movements were analyzed.
The simulated micro-Doppler signature was verified by a measurement
by a X-band (8–12 GHz) radar. Micro-Doppler signature of the human
by UWB radar was presented in [22].
Another solution for detection application is using statistic theories to
test the suggested target against a threshold. For UWB radar detection
applications, methods such as fixed threshold and Constant False Alarm
Rate detectors (CFAR) are proposed [24].
2.3.3 Localization
Electromagnetic (EM) waves propagate in vacuum at the speed of light
(∼ 3 × 108 m/s). The transmitted wave is propagating through some
media that can be the air, cloths or rainfalls. The media affects the
propagation speed of EM waves, but these effects are quite small in
frequencies bellow 10 GHz so they can be disregarded [25]. To calculate
the distance to the target equation 2.3 is used.
Distance to the target = T OA ∗ C0
2
(2.3)
where C0 is the speed of light in vacuum and Time Of Arrival (TOA)
of the detected target is the time that it takes for the electromagnetic
waves to travel from the transmitter to the target and scattered back
again to the receiver.
2.3.4 Tracking
Tracking algorithms are often used to increase the precision of the localization results for moving targets. Most of the tracking algorithms
can make an educated guess about target’s next position and reduce the
measurements uncertainty and smoothen the target trajectory. Kalman
filter, linear least square and particle filter are widely used for this application [26, 27].
2.3 Signal Processing 13
of the human body parts was derived. The time-frequency transform of
the signal was then analyzed to derive the micro-Doppler signature of human movement and complex arm and motion movements were analyzed.
The simulated micro-Doppler signature was verified by a measurement
by a X-band (8–12 GHz) radar. Micro-Doppler signature of the human
by UWB radar was presented in [22].
Another solution for detection application is using statistic theories to
test the suggested target against a threshold. For UWB radar detection
applications, methods such as fixed threshold and Constant False Alarm
Rate detectors (CFAR) are proposed [24].
2.3.3 Localization
Electromagnetic (EM) waves propagate in vacuum at the speed of light
(∼ 3 × 108 m/s). The transmitted wave is propagating through some
media that can be the air, cloths or rainfalls. The media affects the
propagation speed of EM waves, but these effects are quite small in
frequencies bellow 10 GHz so they can be disregarded [25]. To calculate
the distance to the target equation 2.3 is used.
Distance to the target = T OA ∗ C0
2
(2.3)
where C0 is the speed of light in vacuum and Time Of Arrival (TOA)
of the detected target is the time that it takes for the electromagnetic
waves to travel from the transmitter to the target and scattered back
again to the receiver.
2.3.4 Tracking
Tracking algorithms are often used to increase the precision of the localization results for moving targets. Most of the tracking algorithms
can make an educated guess about target’s next position and reduce the
measurements uncertainty and smoothen the target trajectory. Kalman
filter, linear least square and particle filter are widely used for this application [26, 27].
25
14 Chapter 2. Related Work
2.4 Human Radar Cross Section
The term Radar Cross Section (RCS) is defined as the projected area
of a metal sphere that would return the same echo signal as the target.
It depends on many factors such as frequency and polarization of the
incident wave and the target aspect (its orientation relative to the radar
device). Calculation of RCS is a matter of finding the scattered electric field from the target which involves calculating the induced current
on the target by solving Maxwell’s equations for complicated boundary
conditions. This is usually by numerical methods.
For more complex objects such as the human body, the analytical
solution method does not exist. Other approximate methods are used
to estimate the radar cross section of complex objects. RCS can also be
measured by placing the target at radar detection region in free space
or an anechoic chamber [28].
In this thesis an approximate phantom of a human trunk is built to
assist testing and applying simulated conditions for human body electromagnetic field interaction across the frequency band-width of interest
and to make it easier to repeat measurements (Paper C). The RCS of the
phantom is of course measured and compared with RCS of the human.
The human body RCS has been explored by both radar measurements and computer model simulations by several investigators. Dogaru
et al. [29] modelled the radar signature of the human body by Finite Difference Time Domain (FDTD). They used the human body computer
model in various postures in the frequency range of 0.5 GHz to 9 GHz
and all azimuth angles. It was observed that for most frequencies, the
RCS of the body is in a range between −10 and 0 dBsm, where dBsm
is a notation for RCS of a target in decibels. It was also shown that
the posture and amount of fat on the body can affect the RCS, but the
average remains the same for different postures such as standing and
kneeling human. One reason for this is because the main contribution
of the radar reflection is typically coming from the trunk.
Yamada et al. [30] measured the RCS for a human in 76 GHz band.
While the RCS was changing with orientation, the average intensity was
found to be −8.1 dBsm, and as expected, the front and back of the
human trunk produced the largest reflection. It was also shown that the
type of clothing being worn can then affect the radar reflection.
14 Chapter 2. Related Work
2.4 Human Radar Cross Section
The term Radar Cross Section (RCS) is defined as the projected area
of a metal sphere that would return the same echo signal as the target.
It depends on many factors such as frequency and polarization of the
incident wave and the target aspect (its orientation relative to the radar
device). Calculation of RCS is a matter of finding the scattered electric field from the target which involves calculating the induced current
on the target by solving Maxwell’s equations for complicated boundary
conditions. This is usually by numerical methods.
For more complex objects such as the human body, the analytical
solution method does not exist. Other approximate methods are used
to estimate the radar cross section of complex objects. RCS can also be
measured by placing the target at radar detection region in free space
or an anechoic chamber [28].
In this thesis an approximate phantom of a human trunk is built to
assist testing and applying simulated conditions for human body electromagnetic field interaction across the frequency band-width of interest
and to make it easier to repeat measurements (Paper C). The RCS of the
phantom is of course measured and compared with RCS of the human.
The human body RCS has been explored by both radar measurements and computer model simulations by several investigators. Dogaru
et al. [29] modelled the radar signature of the human body by Finite Difference Time Domain (FDTD). They used the human body computer
model in various postures in the frequency range of 0.5 GHz to 9 GHz
and all azimuth angles. It was observed that for most frequencies, the
RCS of the body is in a range between −10 and 0 dBsm, where dBsm
is a notation for RCS of a target in decibels. It was also shown that
the posture and amount of fat on the body can affect the RCS, but the
average remains the same for different postures such as standing and
kneeling human. One reason for this is because the main contribution
of the radar reflection is typically coming from the trunk.
Yamada et al. [30] measured the RCS for a human in 76 GHz band.
While the RCS was changing with orientation, the average intensity was
found to be −8.1 dBsm, and as expected, the front and back of the
human trunk produced the largest reflection. It was also shown that the
type of clothing being worn can then affect the radar reflection.
26
2.5 Related Experimental Systems 15
2.5 Related Experimental Systems
As mention in section 2.1, UWB radar is used for human detection in
LOS or behind obstacles. Furthermore in many UWB applications, the
character of target motion is not known but usually signal processing
algorithms aim to detect either moving persons or static persons. To
the best of my knowledge the only system that combines the moving
and static person detection is presented by Rovaˇnkov´a and Kocur [19].
This thesis aims at detection of humans working in collaboration with
machineries and it is equally important to detect dynamic humans and
static humans.
Experimental UWB systems used for human detection are presented
in this section, and categorized based on if they are used for human
detection in LOS or behind obstacles and whether they aim at detecting
static or dynamic human beings.
2.5.1 Systems for Human Detection in LOS
In this section, the experimental UWB radar systems aiming at detecting
humans in LOS are presented.
Systems for Dynamic Human Detection in LOS
Chang et al. [14] were addressing the problem of providing pedestrian
safety in the presence of moving vehicles by using an UWB pulse-based
mono-static radar. A Specular Multi-Path Model (SMPM) was used
to characterize human body scattered UWB waveforms to detect the
presence of humans via gait. The SMPM was a computationally useful signal representation that reduces UWB waveform representation to
2 dimensions (path amplitude and TOA). First the Moving Target Indication (MTI) system was applied, which rejects highly human-unlike
stationary clutter. Then the CLEAN algorithm [32] was applied to the
MTI response of radar scan to obtain estimated TOAs and amplitudes of
the decomposed multipath components. Thereafter, the signal was segmented in time to isolate the scatters associated with individual moving
objects. Later, each segment was associated to segments from previous
recording intervals with the aid of a Multi-Target Tracking (MTT) technique. A hypothesis testing process determined whether the tested track
was interpreted/detected as a human or not based on three features: (1)
2.5 Related Experimental Systems 15
2.5 Related Experimental Systems
As mention in section 2.1, UWB radar is used for human detection in
LOS or behind obstacles. Furthermore in many UWB applications, the
character of target motion is not known but usually signal processing
algorithms aim to detect either moving persons or static persons. To
the best of my knowledge the only system that combines the moving
and static person detection is presented by Rovaˇnkov´a and Kocur [19].
This thesis aims at detection of humans working in collaboration with
machineries and it is equally important to detect dynamic humans and
static humans.
Experimental UWB systems used for human detection are presented
in this section, and categorized based on if they are used for human
detection in LOS or behind obstacles and whether they aim at detecting
static or dynamic human beings.
2.5.1 Systems for Human Detection in LOS
In this section, the experimental UWB radar systems aiming at detecting
humans in LOS are presented.
Systems for Dynamic Human Detection in LOS
Chang et al. [14] were addressing the problem of providing pedestrian
safety in the presence of moving vehicles by using an UWB pulse-based
mono-static radar. A Specular Multi-Path Model (SMPM) was used
to characterize human body scattered UWB waveforms to detect the
presence of humans via gait. The SMPM was a computationally useful signal representation that reduces UWB waveform representation to
2 dimensions (path amplitude and TOA). First the Moving Target Indication (MTI) system was applied, which rejects highly human-unlike
stationary clutter. Then the CLEAN algorithm [32] was applied to the
MTI response of radar scan to obtain estimated TOAs and amplitudes of
the decomposed multipath components. Thereafter, the signal was segmented in time to isolate the scatters associated with individual moving
objects. Later, each segment was associated to segments from previous
recording intervals with the aid of a Multi-Target Tracking (MTT) technique. A hypothesis testing process determined whether the tested track
was interpreted/detected as a human or not based on three features: (1)
27
16 Chapter 2. Related Work
Table 2.1: State of the art for human being detection with UWB radar, NA stands for Not Applicable.
Author Background
subtraction
Detection LOS/NLOS Hardware Distance Antenna Localization Tracking
Chang et
al. [14]
MTI
testing
hypothseis
MTT & LOS
Domain
Time
TM
0.3
−12.2
m Dipole
Toroidal TOA MTT
Kilic et al.
[2]
Averaging Likelihood
test
LOS Time
DomainTM
1m
−4 m Toroidal
Dipole
criterion
crossing
Threshold NA
Shingu et
al. [3]
NA
amplitude
Maximum LOS
Analyser
Network Max 6 m Horn TOA NA
Kocur [19]
and
Rovaˇnkov´a
averaging
Exponential CFAR, FFT Behind wall Msequence
∼15 m Horn TOA MTT,
Filter
Kalman
Sachs et al.
[4]
filtering
adaptive
High-pass CFAR
rubble
and under
Behind walls
sequence
M- 17 m Spiral CFAR NA
Rane et al.
[5]
subtraction
Range profile
filter &
Band-pass
STA/LTA
peak of
door
and wooden
concrete wall
LOS, behind
Domain
Time
TM
6 m Toroidal
Dipole
STA/LTA NA
Zetik et al.
[6]
eraging
ponential avAdaptive ex- NA Behind wall sequence M- NA Horn TOA NA
et al.
Nezirovic
[7] squares
Linear least- RMD, SVD Pile of bricks
crete pipe
and a con- sequence
M- 1.6 m Horn NA NA
ger [31]
& BuchegOssberger average Removing transform Wavelet Behind wall generator Pulse 1−5 m Horn NA NA
Kocur [8]
and
Rovaˇnkov´a
averaging
Exponential CFAR Behind wall Msequence Max 3 m Horn TOA Kalman Filter
Zhao et al.
[9]
NA HMM
door
and wooden
Gypsum wall
Domain
Time
TM
∼20
−23
m Dipole
Toroidal NA NA
16 Chapter 2. Related Work
Table 2.1: State of the art for human being detection with UWB radar, NA stands for Not Applicable.
Author Background
subtraction
Detection LOS/NLOS Hardware Distance Antenna Localization Tracking
Chang et
al. [14]
MTI
testing
hypothseis
MTT & LOS
Domain
Time
TM
0.3
−12.2
m Dipole
Toroidal TOA MTT
Kilic et al.
[2]
Averaging Likelihood
test
LOS Time
DomainTM
1m
−4 m Toroidal
Dipole
criterion
crossing
Threshold NA
Shingu et
al. [3]
NA
amplitude
Maximum LOS
Analyser
Network Max 6 m Horn TOA NA
Kocur [19]
and
Rovaˇnkov´a
averaging
Exponential CFAR, FFT Behind wall Msequence
∼15 m Horn TOA MTT,
Filter
Kalman
Sachs et al.
[4]
filtering
adaptive
High-pass CFAR
rubble
and under
Behind walls
sequence
M- 17 m Spiral CFAR NA
Rane et al.
[5]
subtraction
Range profile
filter &
Band-pass
STA/LTA
peak of
door
and wooden
concrete wall
LOS, behind
Domain
Time
TM
6 m Toroidal
Dipole
STA/LTA NA
Zetik et al.
[6]
eraging
ponential avAdaptive ex- NA Behind wall sequence M- NA Horn TOA NA
et al.
Nezirovic
[7] squares
Linear least- RMD, SVD Pile of bricks
crete pipe
and a con- sequence
M- 1.6 m Horn NA NA
ger [31]
& BuchegOssberger average Removing transform Wavelet Behind wall generator Pulse 1−5 m Horn NA NA
Kocur [8]
and
Rovaˇnkov´a
averaging
Exponential CFAR Behind wall Msequence Max 3 m Horn TOA Kalman Filter
Zhao et al.
[9]
NA HMM
door
and wooden
Gypsum wall
Domain
Time
TM
∼20
−23
m Dipole
Toroidal NA NA
28
2.5 Related Experimental Systems 17
the paths maximum magnitude, which is relevant to target composition
and cross-section size; (2) the Root Mean Square (RMS) delay spread of
the multipath delay profile (or the RMS range spread), which is relevant
to target size over the range dimension; and (3) the velocity of the target.
This approach was tested in a realistic outdoor environment and shows
better than 80% detection probability with 1.58% false positive. This
system did not provide a solution for static human being detection. The
experiment was done only in scenarios containing moving cars, humans
and fixed objects but the algorithms were not tested when other clutter
such as bicycles and animals were in the detection region of the radar
system.
Bryan et al. [21] used UWB radar to classify eight different human
activities including walking, running, rotating, punching, jumping, transitioning between standing and sitting, crawling and standing still. The
returned signals were processed using Principal Component Analysis
(PCA) so the dimension of the radar data was reduced. A Support
Vector Machine (SVM) was used to classify the activities based on the
signature. The PCA coefficients, target speed, and Fast Fourier Transform (FFT) results of PCA were the inputs to SVM. FFT result of the
PCA coefficient has information about the dominant periodicity of human motion as well as periodicity of micro-motions. A multi-class classification was implemented using a one versus-one method. The accuracy
of the results are reported to be 85%.
Systems for Static Human Detection in LOS
Kilic et al. [2] present a technique for detection of static humans based
on the fact that a human presence induces small low-frequency temporal
variations in the UWB signal that stand out against the background
signal. The experiment was performed in indoor environment where 3
UWB radios from Time Domain were installed as anchor nodes and the
person was standing in 48 different places with 50 cm in between every
position. To remove unwanted background, the averaged signal over time
was subtracted from the raw signal. Later a likelihood test determined
if the human exist in detection region of the radar. In order to locate the
person two different methods were tested: Line Search and Threshold
Crossing. Line search was simply searching the maximum value in the
signal. In some conditions, the largest peaks did not always correspond
to the person, especially if the person was close to strong reflectors. In
2.5 Related Experimental Systems 17
the paths maximum magnitude, which is relevant to target composition
and cross-section size; (2) the Root Mean Square (RMS) delay spread of
the multipath delay profile (or the RMS range spread), which is relevant
to target size over the range dimension; and (3) the velocity of the target.
This approach was tested in a realistic outdoor environment and shows
better than 80% detection probability with 1.58% false positive. This
system did not provide a solution for static human being detection. The
experiment was done only in scenarios containing moving cars, humans
and fixed objects but the algorithms were not tested when other clutter
such as bicycles and animals were in the detection region of the radar
system.
Bryan et al. [21] used UWB radar to classify eight different human
activities including walking, running, rotating, punching, jumping, transitioning between standing and sitting, crawling and standing still. The
returned signals were processed using Principal Component Analysis
(PCA) so the dimension of the radar data was reduced. A Support
Vector Machine (SVM) was used to classify the activities based on the
signature. The PCA coefficients, target speed, and Fast Fourier Transform (FFT) results of PCA were the inputs to SVM. FFT result of the
PCA coefficient has information about the dominant periodicity of human motion as well as periodicity of micro-motions. A multi-class classification was implemented using a one versus-one method. The accuracy
of the results are reported to be 85%.
Systems for Static Human Detection in LOS
Kilic et al. [2] present a technique for detection of static humans based
on the fact that a human presence induces small low-frequency temporal
variations in the UWB signal that stand out against the background
signal. The experiment was performed in indoor environment where 3
UWB radios from Time Domain were installed as anchor nodes and the
person was standing in 48 different places with 50 cm in between every
position. To remove unwanted background, the averaged signal over time
was subtracted from the raw signal. Later a likelihood test determined
if the human exist in detection region of the radar. In order to locate the
person two different methods were tested: Line Search and Threshold
Crossing. Line search was simply searching the maximum value in the
signal. In some conditions, the largest peaks did not always correspond
to the person, especially if the person was close to strong reflectors. In
29
18 Chapter 2. Related Work
this case, not only the paths that were directly reflected off the person,
but also the indirect reflections show fluctuation over time. For that
a threshold-based method was proposed. The results were examined in
terms of missed detection and accuracy. The accuracy was reported to be
less than 65 cm. The authors believe that proposed technique could also
be applicable to cases with multiple humans. However, this experiment
is not tested on moving human beings to validate the method.
Shingu et al. [3] studied human detection and localization with simulation of an UWB radar sensor network. The received power model of
a human body was estimated by Akaike Information Criterion (AIK).
It was concluded that Gaussian distribution was the best fit as a clutter model. The simulation results were compared with real propagation
measurements with a network analyzer. The localization error was shown
to be 14 cm or less.
2.5.2 Systems for Human Detection Behind Obstacles
Human detection behind walls is a dominant subject in the UWB research area. This section starts by presenting systems for both dynamic
and static human detection, followed by systems for only dynamic human
detection, and ending with systems for only detection of static human.
Rovaˇnkov´a and Kocur [19] proposed a method to detect human behind walls by UWB radar. Detection methods for static and dynamic
human beings are quite similar in signal processing procedures. To remove the background, an exponential averaging method was used due to
its robust performance and low complexity. For detection stage for moving clutter the CFAR detector was proposed. TOA is used to localize the
target and Multiple target tracking (MTT) system using a linear Kalman
filtering was used for target tracking. For the static person, in order to
further improve the signal-to-noise ratio of a static target echo, the impulse response with subtracted background was applied to a range filter
which can reduce the clutter residue and noise. Additionally, respiration
as a narrow-band process can also be enhanced by the use of low-pass
filtering. To extract the respiration rate, a horizontal FFT was applied.
A procedure for positioning of a person with unknown or changing motion activity was proposed. The performance of the combined procedure
is demonstrated by an experiment with one static person and one moving person changing his motion activity during the measurement. The
18 Chapter 2. Related Work
this case, not only the paths that were directly reflected off the person,
but also the indirect reflections show fluctuation over time. For that
a threshold-based method was proposed. The results were examined in
terms of missed detection and accuracy. The accuracy was reported to be
less than 65 cm. The authors believe that proposed technique could also
be applicable to cases with multiple humans. However, this experiment
is not tested on moving human beings to validate the method.
Shingu et al. [3] studied human detection and localization with simulation of an UWB radar sensor network. The received power model of
a human body was estimated by Akaike Information Criterion (AIK).
It was concluded that Gaussian distribution was the best fit as a clutter model. The simulation results were compared with real propagation
measurements with a network analyzer. The localization error was shown
to be 14 cm or less.
2.5.2 Systems for Human Detection Behind Obstacles
Human detection behind walls is a dominant subject in the UWB research area. This section starts by presenting systems for both dynamic
and static human detection, followed by systems for only dynamic human
detection, and ending with systems for only detection of static human.
Rovaˇnkov´a and Kocur [19] proposed a method to detect human behind walls by UWB radar. Detection methods for static and dynamic
human beings are quite similar in signal processing procedures. To remove the background, an exponential averaging method was used due to
its robust performance and low complexity. For detection stage for moving clutter the CFAR detector was proposed. TOA is used to localize the
target and Multiple target tracking (MTT) system using a linear Kalman
filtering was used for target tracking. For the static person, in order to
further improve the signal-to-noise ratio of a static target echo, the impulse response with subtracted background was applied to a range filter
which can reduce the clutter residue and noise. Additionally, respiration
as a narrow-band process can also be enhanced by the use of low-pass
filtering. To extract the respiration rate, a horizontal FFT was applied.
A procedure for positioning of a person with unknown or changing motion activity was proposed. The performance of the combined procedure
is demonstrated by an experiment with one static person and one moving person changing his motion activity during the measurement. The
30
2.5 Related Experimental Systems 19
results showed that it is possible to obtain more extensive and precise
information about the position of monitored persons at the expense of
computational complexity and time consumption. The presented timing
constraint of this system (75 s to detect respiration activity) implies that
this system is not suitable for time critical safety applications.
Sachs et al. [4] used a M-sequence UWB radar with 3-antenna configuration (one transmitter and two receiver antennas) to detect trapped
humans under rubble and moving humans. For moving humans detection, the first signal processing step consists of removing the clutter
caused from static objects by high-pass adaptive filtering. In the second step, CFAR processing for target indication and the corresponding
TOA. Outliers and missing points were then removed and finally, the target trajectory was calculated. The paper concluded that the 3-antenna
configuration is not robust against ghost targets by its principle. To
detect trapped humans under rubble, the small movements such as respiration was out of interest. The signal was enhanced by narrow-band
filtering carried out by applying a horizontal FFT on the data after
clutter removal. Additionally, a sliding average with a short window
was performed over the transformed signal in the propagation time direction. The experiment could detect the breathing person lying in a
cement pipe. The paper presented the results of static and dynamic
humans separately to prove that it was possible to measure both cases
with the mentioned hardware, but it did not provide any information
if it was possible to combine these methods for detection of static and
dynamic human beings at the same time.
Systems for Dynamic Human Detection Behind Obstacles
Rane et al. proposed a Short Term to Long Term Average ratio (STA/LTA)
to detect moving target by Time Domain UWB radar in LOS and behind obstacles [5]. The method is based on enveloping the signal by the
Hilbert transform and averaging. The results showed that this method
shows 79.6% more accuracy than choosing the simple peak of processed
range profile to estimate the distance. For pre-processing the Infinite
Impulse Response (IIR) bandpass filter of order 20 and passband of 3.1–
5.3 GHz is applied along fast time dimension to suppress frequencies
beyond the radar operating range. Stationary clutter is suppressed by
performing weighted Range Profile Subtraction (RPS) on each range
profile. This paper did not consider stationary human being or multiple
2.5 Related Experimental Systems 19
results showed that it is possible to obtain more extensive and precise
information about the position of monitored persons at the expense of
computational complexity and time consumption. The presented timing
constraint of this system (75 s to detect respiration activity) implies that
this system is not suitable for time critical safety applications.
Sachs et al. [4] used a M-sequence UWB radar with 3-antenna configuration (one transmitter and two receiver antennas) to detect trapped
humans under rubble and moving humans. For moving humans detection, the first signal processing step consists of removing the clutter
caused from static objects by high-pass adaptive filtering. In the second step, CFAR processing for target indication and the corresponding
TOA. Outliers and missing points were then removed and finally, the target trajectory was calculated. The paper concluded that the 3-antenna
configuration is not robust against ghost targets by its principle. To
detect trapped humans under rubble, the small movements such as respiration was out of interest. The signal was enhanced by narrow-band
filtering carried out by applying a horizontal FFT on the data after
clutter removal. Additionally, a sliding average with a short window
was performed over the transformed signal in the propagation time direction. The experiment could detect the breathing person lying in a
cement pipe. The paper presented the results of static and dynamic
humans separately to prove that it was possible to measure both cases
with the mentioned hardware, but it did not provide any information
if it was possible to combine these methods for detection of static and
dynamic human beings at the same time.
Systems for Dynamic Human Detection Behind Obstacles
Rane et al. proposed a Short Term to Long Term Average ratio (STA/LTA)
to detect moving target by Time Domain UWB radar in LOS and behind obstacles [5]. The method is based on enveloping the signal by the
Hilbert transform and averaging. The results showed that this method
shows 79.6% more accuracy than choosing the simple peak of processed
range profile to estimate the distance. For pre-processing the Infinite
Impulse Response (IIR) bandpass filter of order 20 and passband of 3.1–
5.3 GHz is applied along fast time dimension to suppress frequencies
beyond the radar operating range. Stationary clutter is suppressed by
performing weighted Range Profile Subtraction (RPS) on each range
profile. This paper did not consider stationary human being or multiple
31
20 Chapter 2. Related Work
target detection.
Zetik et al. [6] described the architecture and design of an M-sequence
UWB through-the-wall radar for detection and localization of moving
humans. To remove clutter an adaptive background subtraction method
was proposed. This method is based on the exponential averaging although instead of a scalar weighing factor α, a vector of weighing timevariant coefficients αk was introduced. The location of the target was
measured by TOA. The achieved precision is in an order of 40 cm. This
paper did not consider stationary human being or multiple target detection.
In [33], Ram et al. used micro-Doppler processing to detect humans
behind obstacle. The simulation showed that the wall introduces minor
distortion on micro-Doppler effect of human body parts but it affects the
magnitude response of Doppler effect in form of attenuation and fading.
The experiment was based on a Doppler radar, but the preliminary investigations by the authors showed that the same methodology could be
applied to a wide-band radar.
Systems for Static Human Detection Behind Obstacles
Nezirovi´c et al. [7] presented a Respiratory Motion Detection (RMD)
algorithm. The algorithm was based on RMD algorithm presented in [4]
and complementing it with SVD. The purpose of applying SVD is to
separate the respiratory motion response from the non-stationary clutter, and in addition, reduce the effect of the Additive White Gaussian
Noise (AWGN). A linear least-squares method was proposed to remove
stationary clutter along with any potential linear trend in the slow-time
dimension. The results compared the developed RMD with the reference RMD. The performance of the RMD algorithm was assessed both
by means of a Monte Carlo simulation as well as experimental data acquired under realistic conditions of a person lying in concrete pipe under
1.2 meter bricks and moist soil. The results showed increase in performance of the developed RMD algorithm over the reference RMD algorithm in terms of its de-noising capabilities and non-stationary clutter
separation.
Ossberger et al. [31] presented a Continuous Wavelet Transform (CWT)
for detection of respiratory activity thorough wall with an UWB pulse
radar test set-up. CWT is suitable for non-stationary signals like pulses.
It was shown that the UWB radar can detect respiration at up to 5
20 Chapter 2. Related Work
target detection.
Zetik et al. [6] described the architecture and design of an M-sequence
UWB through-the-wall radar for detection and localization of moving
humans. To remove clutter an adaptive background subtraction method
was proposed. This method is based on the exponential averaging although instead of a scalar weighing factor α, a vector of weighing timevariant coefficients αk was introduced. The location of the target was
measured by TOA. The achieved precision is in an order of 40 cm. This
paper did not consider stationary human being or multiple target detection.
In [33], Ram et al. used micro-Doppler processing to detect humans
behind obstacle. The simulation showed that the wall introduces minor
distortion on micro-Doppler effect of human body parts but it affects the
magnitude response of Doppler effect in form of attenuation and fading.
The experiment was based on a Doppler radar, but the preliminary investigations by the authors showed that the same methodology could be
applied to a wide-band radar.
Systems for Static Human Detection Behind Obstacles
Nezirovi´c et al. [7] presented a Respiratory Motion Detection (RMD)
algorithm. The algorithm was based on RMD algorithm presented in [4]
and complementing it with SVD. The purpose of applying SVD is to
separate the respiratory motion response from the non-stationary clutter, and in addition, reduce the effect of the Additive White Gaussian
Noise (AWGN). A linear least-squares method was proposed to remove
stationary clutter along with any potential linear trend in the slow-time
dimension. The results compared the developed RMD with the reference RMD. The performance of the RMD algorithm was assessed both
by means of a Monte Carlo simulation as well as experimental data acquired under realistic conditions of a person lying in concrete pipe under
1.2 meter bricks and moist soil. The results showed increase in performance of the developed RMD algorithm over the reference RMD algorithm in terms of its de-noising capabilities and non-stationary clutter
separation.
Ossberger et al. [31] presented a Continuous Wavelet Transform (CWT)
for detection of respiratory activity thorough wall with an UWB pulse
radar test set-up. CWT is suitable for non-stationary signals like pulses.
It was shown that the UWB radar can detect respiration at up to 5
32
2.5 Related Experimental Systems 21
meters.
Zhao, Liang, and Durrani [9] used Hidden Markov Model (HMM) to
classify the target in presence of clutter. The experiment was done in
two scenarios: a scenario with dense foliage and a 1.5 meter 90° corner
metal reflector and a through wall scenario with a human target. The
paper presented the detection results in terms of false positive and negative outputs. Results showed the possibility of using HMM for target
detection. For through-wall detection scenario, HMMs also showed good
capability to distinguish between radar signals containing human target
and no target.
2.5 Related Experimental Systems 21
meters.
Zhao, Liang, and Durrani [9] used Hidden Markov Model (HMM) to
classify the target in presence of clutter. The experiment was done in
two scenarios: a scenario with dense foliage and a 1.5 meter 90° corner
metal reflector and a through wall scenario with a human target. The
paper presented the detection results in terms of false positive and negative outputs. Results showed the possibility of using HMM for target
detection. For through-wall detection scenario, HMMs also showed good
capability to distinguish between radar signals containing human target
and no target.
33
Chapter 3. System Design and Validation
Chapter 3
System Design and
Validation
As presented in chapter 2, most UWB radar systems for human detection
applications are developed for surveillance and rescue of humans behind
obstacles, and are focused on either static or dynamic human detection.
These systems generally do not consider the effect of environmental noise
or different postures, physical builds or size for different humans on signal
detection, and do not discuss the shortcomings of UWB radar in human
detection applications. In this section some of aforementioned issues are
addressed and the contributions and results achieved within this thesis
are presented.
This section contains three parts: Part one presents the design of
a data acquisition application. Part two presents the validation of the
radar system through measurements. These measurements were performed to evaluate the radar system’s capabilities and shortcomings. In
the third part some of the signal processing algorithms that are developed for detection of humans with UWB radar are presented. The results
and discussions for every part are presented.
3.1 Software Platform Design
The Radarbolaget’s UWB radar system is delivered with a windows application for data acquisition and processing. The drawback of Radarbo22
Chapter 3. System Design and Validation
Chapter 3
System Design and
Validation
As presented in chapter 2, most UWB radar systems for human detection
applications are developed for surveillance and rescue of humans behind
obstacles, and are focused on either static or dynamic human detection.
These systems generally do not consider the effect of environmental noise
or different postures, physical builds or size for different humans on signal
detection, and do not discuss the shortcomings of UWB radar in human
detection applications. In this section some of aforementioned issues are
addressed and the contributions and results achieved within this thesis
are presented.
This section contains three parts: Part one presents the design of
a data acquisition application. Part two presents the validation of the
radar system through measurements. These measurements were performed to evaluate the radar system’s capabilities and shortcomings. In
the third part some of the signal processing algorithms that are developed for detection of humans with UWB radar are presented. The results
and discussions for every part are presented.
3.1 Software Platform Design
The Radarbolaget’s UWB radar system is delivered with a windows application for data acquisition and processing. The drawback of Radarbo22
34
3.2 System Validation and Measurements 23
Figure 3.1: Data acquisition and processing system.
laget’s windows application is that to test a new algorithm on live radar
data, the algorithm needs to be coded in C which is a time consuming task. To overcome this, a windows application for data acquisition
and processing is designed and implemented by a team from Addiva AB
including the author of this thesis. The idea behind designing this application is to create a platform enabling direct use of MATLAB algorithms
and libraries for signal processing.
The application platform is built in a way that the algorithm parameters and their sequence is editable in the user interface. By using
MATLAB we have access to the advanced and computationally enhanced
signal processing algorithms, which can be applied directly on the live
and raw data. An overview of the system is shown in Figure 3.1. A
screenshot of the application graphical user interface is shown in Figure
3.2.
3.2 System Validation and Measurements
Several measurements are planned and performed in order to evaluate
the UWB radar system and to develop signal processing algorithms able
to detect and localize both static and dynamic humans in enclosed environments, which is the goal of this thesis.
In this section measurements that are performed to validate the constraints, characteristics and abilities of the system are presented.
3.2 System Validation and Measurements 23
Figure 3.1: Data acquisition and processing system.
laget’s windows application is that to test a new algorithm on live radar
data, the algorithm needs to be coded in C which is a time consuming task. To overcome this, a windows application for data acquisition
and processing is designed and implemented by a team from Addiva AB
including the author of this thesis. The idea behind designing this application is to create a platform enabling direct use of MATLAB algorithms
and libraries for signal processing.
The application platform is built in a way that the algorithm parameters and their sequence is editable in the user interface. By using
MATLAB we have access to the advanced and computationally enhanced
signal processing algorithms, which can be applied directly on the live
and raw data. An overview of the system is shown in Figure 3.1. A
screenshot of the application graphical user interface is shown in Figure
3.2.
3.2 System Validation and Measurements
Several measurements are planned and performed in order to evaluate
the UWB radar system and to develop signal processing algorithms able
to detect and localize both static and dynamic humans in enclosed environments, which is the goal of this thesis.
In this section measurements that are performed to validate the constraints, characteristics and abilities of the system are presented.
35
24 Chapter 3. System Design and Validation Figure 3.2: Data acquisition and processing application. The raw and processed radar data from a recorded and the input parameters and order of them can be modified. measurement is shown in the middle frame. In the right frame the signal processing algorithms can be chosen
24 Chapter 3. System Design and Validation Figure 3.2: Data acquisition and processing application. The raw and processed radar data from a recorded and the input parameters and order of them can be modified. measurement is shown in the middle frame. In the right frame the signal processing algorithms can be chosen
36
3.2 System Validation and Measurements 25
Figure 3.3: Measurements of Vertical and horizontal polarization of
Vivaldi antennas for 1 GHz, 2 GHz and 3 GHz.
3.2.1 The Vivaldi Antenna Radiation Pattern Measurement
In this section, an overview of measurement of the Vivaldi antenna radiation pattern is presented. The antenna radiation pattern is an important
factor in radar measurements in order to understand the radar interaction with the object or a human under test.
The Vivaldi antennas radiation pattern is measured in a semi anechoic chamber at DELTA Development Technology by the great-circle
method [34]. The floor in the semi anechoic chamber does not have
absorbing tiles therefore the antennas are positioned on 3 m height to
reduce the reflection from the floor. In the great-circle method the Measurement Antenna (MA), which in this set-up is a Horn antenna, is
fixed and the Antenna Under Test (AUT), which is the Vivaldi antenna,
is placed on a rotational positioner and rotated around the azimuth
through 360° to generate a two-dimensional polar pattern. The rotational positioner is rotating in 5° each steps. The same measurement is
repeated when the Vivaldi antenna is manually rotated 90° to measure
the horizontal polarization. The measurement set-up is shown in Figure
3.2 System Validation and Measurements 25
Figure 3.3: Measurements of Vertical and horizontal polarization of
Vivaldi antennas for 1 GHz, 2 GHz and 3 GHz.
3.2.1 The Vivaldi Antenna Radiation Pattern Measurement
In this section, an overview of measurement of the Vivaldi antenna radiation pattern is presented. The antenna radiation pattern is an important
factor in radar measurements in order to understand the radar interaction with the object or a human under test.
The Vivaldi antennas radiation pattern is measured in a semi anechoic chamber at DELTA Development Technology by the great-circle
method [34]. The floor in the semi anechoic chamber does not have
absorbing tiles therefore the antennas are positioned on 3 m height to
reduce the reflection from the floor. In the great-circle method the Measurement Antenna (MA), which in this set-up is a Horn antenna, is
fixed and the Antenna Under Test (AUT), which is the Vivaldi antenna,
is placed on a rotational positioner and rotated around the azimuth
through 360° to generate a two-dimensional polar pattern. The rotational positioner is rotating in 5° each steps. The same measurement is
repeated when the Vivaldi antenna is manually rotated 90° to measure
the horizontal polarization. The measurement set-up is shown in Figure
37
26 Chapter 3. System Design and Validation
Figure 3.4: The Vivaldi antenna radiation pattern measurement set-up
in a semi-anechoic chamber.
Figure 3.5: The Vivaldi antenna connected to a coaxial cable with a
balun and installed on a rotational positioner.
26 Chapter 3. System Design and Validation
Figure 3.4: The Vivaldi antenna radiation pattern measurement set-up
in a semi-anechoic chamber.
Figure 3.5: The Vivaldi antenna connected to a coaxial cable with a
balun and installed on a rotational positioner.
38
3.2 System Validation and Measurements 27
3.4 and Figure 3.5.
A balun with 3 dB damping following by a pair of semi rigid coaxial
cables connects the Vivaldi antenna to a synthesized sweeper that generates CW waves from 1–3 GHz with the frequency step size of 1 GHz.
In this measurement setup, the horn antenna with an built-in amplifier is connected to an Electromagnetic Interference (EMI) test receiver.
The results for both horizontal and vertical polarization measurements
are shown in Figure 3.3. The results show that there is no significant
directivity at 1 GHz. At 2 and 3 GHz the directivity is quite acceptable.
3.2.2 Comparison of M-sequence vs Pulse radar for
Static Human Detection (Paper A)
As presented in section 2, the majority of UWB radars systems used for
human detection are either pulse radar from Time DomainTM (currently
Humatics 1
) or M-sequence. These two systems are compared from hardware prospective and expected performance in [35]. To compare these
two systems performances, measurements were performed in a university
corridor with a static human standing in LOS at 4, 8 and 12 m respectively. The results are presented in paper (A). The better performance of
the M-sequence is due to the fact that the reflected wave from the target
correlates with transmitted M-sequence, which is a known code sequence.
This process causes M-sequence to have higher power consumption than
pulse radar but reduces the reflection from previous radar sweeps and
other sources of noise. To the best of my knowledge this is the First time
these two radar systems are compared in an experimental set-up.
This measurement was performed in order to analyse the throughput
of the radar system for static human detection and is partly answering
RQ1.
3.2.3 Walking Human Detection in Different Environments (Paper B)
One of the requirements for the system used in thesis is that the system shall be able to perform in a cluttered environment where moving
machineries and vehicles are working in collaboration with human beings. Additionally the proposed system requires a short response time,
1https://www.humatics.com/
3.2 System Validation and Measurements 27
3.4 and Figure 3.5.
A balun with 3 dB damping following by a pair of semi rigid coaxial
cables connects the Vivaldi antenna to a synthesized sweeper that generates CW waves from 1–3 GHz with the frequency step size of 1 GHz.
In this measurement setup, the horn antenna with an built-in amplifier is connected to an Electromagnetic Interference (EMI) test receiver.
The results for both horizontal and vertical polarization measurements
are shown in Figure 3.3. The results show that there is no significant
directivity at 1 GHz. At 2 and 3 GHz the directivity is quite acceptable.
3.2.2 Comparison of M-sequence vs Pulse radar for
Static Human Detection (Paper A)
As presented in section 2, the majority of UWB radars systems used for
human detection are either pulse radar from Time DomainTM (currently
Humatics 1
) or M-sequence. These two systems are compared from hardware prospective and expected performance in [35]. To compare these
two systems performances, measurements were performed in a university
corridor with a static human standing in LOS at 4, 8 and 12 m respectively. The results are presented in paper (A). The better performance of
the M-sequence is due to the fact that the reflected wave from the target
correlates with transmitted M-sequence, which is a known code sequence.
This process causes M-sequence to have higher power consumption than
pulse radar but reduces the reflection from previous radar sweeps and
other sources of noise. To the best of my knowledge this is the First time
these two radar systems are compared in an experimental set-up.
This measurement was performed in order to analyse the throughput
of the radar system for static human detection and is partly answering
RQ1.
3.2.3 Walking Human Detection in Different Environments (Paper B)
One of the requirements for the system used in thesis is that the system shall be able to perform in a cluttered environment where moving
machineries and vehicles are working in collaboration with human beings. Additionally the proposed system requires a short response time,
1https://www.humatics.com/
39
28 Chapter 3. System Design and Validation
so when the risk for collision arises it becomes possible to stop the vehicle
in advance. The reliability of the system is also an important factor that
can be presented in values such as false positive and negative outputs.
In paper (B) the results of an experimental comparison study of
human movements and presence detection in two different environments,
a semi-anechoic chamber and open office, using UWB M-Sequence radar
is presented. The signal processing algorithms presented in the paper
are able to track the target movements.
The results presented in paper (B) show that the cluttered environment has a direct effect on the false positive mainly due to multipath
reflections. The false negative outputs are mostly caused by the delay
in the human detection algorithm. This measurement was planned for
analyzing the throughput of a radar system for dynamic human detection and is partly answering the RQ1 by presenting the results of clutter
distribution of radar signal in different environments. The signal processing algorithms and measurements of the false positive and negative
outputs are partly answering RQ2.
3.2.4 Respiration Simulation and Measurement
One of the requirements for the system used in thesis is that the system
shall be able to detect dynamic humans as well as static humans. Respiratory activity was used as an indicator for presence of static humans in
literature presented in chapter 2. During the respiration of an average
adult human, the chest displacement is around 3 mm and the respiratory
rate corresponds to 12 – 20 breaths per minute.
The good spatial resolution of UWB radar creates the possibility of
detection of chest movements in humans. As presented in chapter 2,
different processing techniques such as narrow-band filtering followed by
FFT [4, 19], Wavelet transform [31] and micro-Doppler processing [33]
are used for detection of the respiratory activity.
In this thesis, the movement of the chest is simulated with the periodic movement of an aluminium plate, which should be easier to detect
for the radar than the measurement on a real human. In this measurement, the antenna pair, sender and receiver, are placed on a stand and
an aluminium plate of the size 20 × 20 cm2
is moved periodically for 4
cm and 90 degrees to bore sight of the antennas. The set-up is shown
in Figure 3.6. The xy board used for this set-up is built by a team at
Addiva AB including the author of this thesis.
28 Chapter 3. System Design and Validation
so when the risk for collision arises it becomes possible to stop the vehicle
in advance. The reliability of the system is also an important factor that
can be presented in values such as false positive and negative outputs.
In paper (B) the results of an experimental comparison study of
human movements and presence detection in two different environments,
a semi-anechoic chamber and open office, using UWB M-Sequence radar
is presented. The signal processing algorithms presented in the paper
are able to track the target movements.
The results presented in paper (B) show that the cluttered environment has a direct effect on the false positive mainly due to multipath
reflections. The false negative outputs are mostly caused by the delay
in the human detection algorithm. This measurement was planned for
analyzing the throughput of a radar system for dynamic human detection and is partly answering the RQ1 by presenting the results of clutter
distribution of radar signal in different environments. The signal processing algorithms and measurements of the false positive and negative
outputs are partly answering RQ2.
3.2.4 Respiration Simulation and Measurement
One of the requirements for the system used in thesis is that the system
shall be able to detect dynamic humans as well as static humans. Respiratory activity was used as an indicator for presence of static humans in
literature presented in chapter 2. During the respiration of an average
adult human, the chest displacement is around 3 mm and the respiratory
rate corresponds to 12 – 20 breaths per minute.
The good spatial resolution of UWB radar creates the possibility of
detection of chest movements in humans. As presented in chapter 2,
different processing techniques such as narrow-band filtering followed by
FFT [4, 19], Wavelet transform [31] and micro-Doppler processing [33]
are used for detection of the respiratory activity.
In this thesis, the movement of the chest is simulated with the periodic movement of an aluminium plate, which should be easier to detect
for the radar than the measurement on a real human. In this measurement, the antenna pair, sender and receiver, are placed on a stand and
an aluminium plate of the size 20 × 20 cm2
is moved periodically for 4
cm and 90 degrees to bore sight of the antennas. The set-up is shown
in Figure 3.6. The xy board used for this set-up is built by a team at
Addiva AB including the author of this thesis.
40
3.2 System Validation and Measurements 29
Figure 3.6: Respiration Simulation Set-up. An aluminium plate moving periodically by a stepper motors 90° to the bore sight of the antennas.
The result is shown in Figure 3.7. The antenna cross talk was removed from the measurement data and the data was up-sampled (see
3.3.5) to increase the accuracy. The periodic movement of the aluminium
plate is clearly visible in the radargram at around 3 m which is twice of
the distance of aluminium plate to the antennas.
The measurement is repeated with a real human sitting in-front of
the radar and breathing normally at Addiva AB office by the author of
this thesis. Some algorithms such as narrow-band filtering and FFT are
tested on the signal to detect respiration activity but it did not produce
satisfactory results. This might be related to fluctuation in signal at
every radar sweep or it may be because the algorithm is not suited for
this particular hardware. This measurement is partly answering RQ1
and RQ2.
3.2.5 Phantom Measurements (Paper C)
In paper (C), the measurement of a human torso phantom is presented.
Humans are complicated targets because of the complex surface profile
and difference in clothing and body size. A simplified phantom of the
human torso can replicate a human the in radar measurements. A circular cylinder pipe made of PVC with approximate dimension of a human
trunk is chosen as a phantom. The phantom is filled with a mixture of
sand, water and salt that creates a similar permittivity as the human
3.2 System Validation and Measurements 29
Figure 3.6: Respiration Simulation Set-up. An aluminium plate moving periodically by a stepper motors 90° to the bore sight of the antennas.
The result is shown in Figure 3.7. The antenna cross talk was removed from the measurement data and the data was up-sampled (see
3.3.5) to increase the accuracy. The periodic movement of the aluminium
plate is clearly visible in the radargram at around 3 m which is twice of
the distance of aluminium plate to the antennas.
The measurement is repeated with a real human sitting in-front of
the radar and breathing normally at Addiva AB office by the author of
this thesis. Some algorithms such as narrow-band filtering and FFT are
tested on the signal to detect respiration activity but it did not produce
satisfactory results. This might be related to fluctuation in signal at
every radar sweep or it may be because the algorithm is not suited for
this particular hardware. This measurement is partly answering RQ1
and RQ2.
3.2.5 Phantom Measurements (Paper C)
In paper (C), the measurement of a human torso phantom is presented.
Humans are complicated targets because of the complex surface profile
and difference in clothing and body size. A simplified phantom of the
human torso can replicate a human the in radar measurements. A circular cylinder pipe made of PVC with approximate dimension of a human
trunk is chosen as a phantom. The phantom is filled with a mixture of
sand, water and salt that creates a similar permittivity as the human
41
30 Chapter 3. System Design and Validation
Figure 3.7: Respiration Simulation Results. The periodic movement of
the aluminium plate is clearly visible from 20 to 60 s in the radargram
at around 3 m which is twice of the distance of aluminium plate to the
antennas. The antenna cross talk was removed from the measurement
data.
body. The RCS of the empty pipe, filled pipe and a human sitting in
front of the radar is measured and compared. The results show that
it is possible to distinguish the distances between human breast, neck
and head in the human RCS and the empty pipe front and back in the
empty pipe RCS. The RCS of filled pipe has higher amplitude than the
human but the overall summed signal is similar. The RCS of an metal
aluminium plate is also measured in different polarization of the transmitter antenna. This measurement and its results are answering RQ3.
In Paper (C), the frequency choice for the radar system for discerning
human trunk based on Mie theory is discussed. It is concluded that an
object with the size and geometry of human torso can have larger cross
section if the frequency of the radar system is lower than the frequency
band in the existing system but a system with lower frequency might
not provide the same spatial resolution. This is partly answering RQ1.
30 Chapter 3. System Design and Validation
Figure 3.7: Respiration Simulation Results. The periodic movement of
the aluminium plate is clearly visible from 20 to 60 s in the radargram
at around 3 m which is twice of the distance of aluminium plate to the
antennas. The antenna cross talk was removed from the measurement
data.
body. The RCS of the empty pipe, filled pipe and a human sitting in
front of the radar is measured and compared. The results show that
it is possible to distinguish the distances between human breast, neck
and head in the human RCS and the empty pipe front and back in the
empty pipe RCS. The RCS of filled pipe has higher amplitude than the
human but the overall summed signal is similar. The RCS of an metal
aluminium plate is also measured in different polarization of the transmitter antenna. This measurement and its results are answering RQ3.
In Paper (C), the frequency choice for the radar system for discerning
human trunk based on Mie theory is discussed. It is concluded that an
object with the size and geometry of human torso can have larger cross
section if the frequency of the radar system is lower than the frequency
band in the existing system but a system with lower frequency might
not provide the same spatial resolution. This is partly answering RQ1.
42
3.3 Signal Processing Algorithms 31
3.3 Signal Processing Algorithms
In this thesis, some of the algorithms addressed in the related work literature are implemented and tested on the obtained signals from measurements. Some other algorithms are developed by empirical trials. These
algorithms are presented in this chapter and are partly answering RQ2.
3.3.1 Pre-processing
The main objective for pre-processing is to remove noise from the raw
signal.
• Hilbert Transform: Hilbert transform of function u(t) is defined as
the following function:
H(u)(t) = 1
π
Z ∞
−∞
u(τ )
t − τ
dτ (3.1)
This linear operator is given by convolution with the function
1/(πt). The Hilbert transform H(u)(t) in the frequency domain
applies a phase shift of 90° to every Fourier component of a function. Therefore, H(u)(t) has the effect of shifting the phase of the
negative frequency components of u(t) by +90° (π/2 radians) and
the phase of the positive frequency components by 90°. Hilbert
transform is resulting in enveloping the signal, reducing the noise,
and make it easier to find the peaks that represent the target. The
result of the transform is shown in Figure 3.8.
In [5], Rane et al. show that by using Hilbert transform and averaging instead of choosing a simple peak to estimate the distance to
the target the accuracy increases. This performance improvement
is not investigated in this thesis.
• Automatic noise reduction: This filter sets a limit based on the
standard deviation of the signal calculated over the corresponding
data points of all radar sweeps (slow time) multiply by a sensitivity
factor. The sensitivity is an input argument to the automatic noise
reduction function. Any data point with an absolute value less than
that multiplication value is replaced with zero.
This filter removes the signal values with small amplitude from the
signal based on the desired sensitivity and reduces the probability
3.3 Signal Processing Algorithms 31
3.3 Signal Processing Algorithms
In this thesis, some of the algorithms addressed in the related work literature are implemented and tested on the obtained signals from measurements. Some other algorithms are developed by empirical trials. These
algorithms are presented in this chapter and are partly answering RQ2.
3.3.1 Pre-processing
The main objective for pre-processing is to remove noise from the raw
signal.
• Hilbert Transform: Hilbert transform of function u(t) is defined as
the following function:
H(u)(t) = 1
π
Z ∞
−∞
u(τ )
t − τ
dτ (3.1)
This linear operator is given by convolution with the function
1/(πt). The Hilbert transform H(u)(t) in the frequency domain
applies a phase shift of 90° to every Fourier component of a function. Therefore, H(u)(t) has the effect of shifting the phase of the
negative frequency components of u(t) by +90° (π/2 radians) and
the phase of the positive frequency components by 90°. Hilbert
transform is resulting in enveloping the signal, reducing the noise,
and make it easier to find the peaks that represent the target. The
result of the transform is shown in Figure 3.8.
In [5], Rane et al. show that by using Hilbert transform and averaging instead of choosing a simple peak to estimate the distance to
the target the accuracy increases. This performance improvement
is not investigated in this thesis.
• Automatic noise reduction: This filter sets a limit based on the
standard deviation of the signal calculated over the corresponding
data points of all radar sweeps (slow time) multiply by a sensitivity
factor. The sensitivity is an input argument to the automatic noise
reduction function. Any data point with an absolute value less than
that multiplication value is replaced with zero.
This filter removes the signal values with small amplitude from the
signal based on the desired sensitivity and reduces the probability
43
32 Chapter 3. System Design and Validation
Figure 3.8: Absolute value of the Hilbert transform of the signal.
of false positive by reducing the probability of the small amplitude
noise being mistaken for targets. This filter is suitable for offline
processing of the signal, when the measurement is completed and
there is an access to all measured radar returns. It is implemented
and applied on the measurement data with good results.
3.3.2 Clutter Removal
As previously mentioned in section 2.3.1 clutter is an unwanted radar
return signal. In this thesis, the aim of clutter removal is to detect the
radar response of static and dynamic humans in the detection region of
the radar. In UWB radar literature, the term “background” refers to the
clutter which contains, among others, the antenna cross-talk and waves
reflected from static objects except the human in the scenario [6, 19].
In this thesis, to remove background from the radar data two approaches are tested: In case of an static object, a reference measurement
is performed before the object of interest is placed in the radar detection
region. The obtained references measurement is processed by methods
such as averaging, removal of the absolute maximum, and median re32 Chapter 3. System Design and Validation
Figure 3.8: Absolute value of the Hilbert transform of the signal.
of false positive by reducing the probability of the small amplitude
noise being mistaken for targets. This filter is suitable for offline
processing of the signal, when the measurement is completed and
there is an access to all measured radar returns. It is implemented
and applied on the measurement data with good results.
3.3.2 Clutter Removal
As previously mentioned in section 2.3.1 clutter is an unwanted radar
return signal. In this thesis, the aim of clutter removal is to detect the
radar response of static and dynamic humans in the detection region of
the radar. In UWB radar literature, the term “background” refers to the
clutter which contains, among others, the antenna cross-talk and waves
reflected from static objects except the human in the scenario [6, 19].
In this thesis, to remove background from the radar data two approaches are tested: In case of an static object, a reference measurement
is performed before the object of interest is placed in the radar detection
region. The obtained references measurement is processed by methods
such as averaging, removal of the absolute maximum, and median re44
3.3 Signal Processing Algorithms 33
moval. This approach is suitable in a measurement scenario when there
is the possibility to measure the radar response of a static background
before the object of interest is placed in the radar detection region.
For dynamic objects and environments, methods such as exponential
averaging and adaptive background subtraction is used. Methods such
as averaging, removal of the absolute maximum, and median removal
can also be used for background removal of dynamic object detection
but it is only suitable for offline processing when the measurement is
completed and there is an access to all measured data. These methods
are presented in the following:
• Averaging: This method makes the reference frame simply by averaging corresponding data values in reference measurement (slow
time, see 2.4) and removing it from every frame in the measurement data. This method is widely used in UWB radar background
processing [2,31]. The experimental results in this thesis show that
this technique effectively removes the clutters.
• Removal of absolute maximum: In this method the absolute maximum of corresponding data points of all data in the reference measurement (slow time) is founded. This method can help removing
high peaks showing up in some frames due to the noise. The disadvantage of this method is that it might remove some important
data with high amplitude. This method was developed based on
empirical trials and showed a good performance when an activity
or movement was to be detected in a cluttered environment.
• Median removal: In this method the median of corresponding data
points of all data in the reference measurement (slow time) is removed from the measurement frames. This method was developed
based on empirical trials and showed a good performance when an
activity or movement was to be detected in a cluttered environment.
• Exponential averaging: Exponential averaging algorithm, computes
a new background estimate yk from the previous background yk−1
that is updated by a new measured radar return xk
yk = αyk−1 + (1 − α)xk (3.2)
where α is the constant scalar weighing factor with the value between 0 and 1, yk and xk are vectors with the size of [1× m]
3.3 Signal Processing Algorithms 33
moval. This approach is suitable in a measurement scenario when there
is the possibility to measure the radar response of a static background
before the object of interest is placed in the radar detection region.
For dynamic objects and environments, methods such as exponential
averaging and adaptive background subtraction is used. Methods such
as averaging, removal of the absolute maximum, and median removal
can also be used for background removal of dynamic object detection
but it is only suitable for offline processing when the measurement is
completed and there is an access to all measured data. These methods
are presented in the following:
• Averaging: This method makes the reference frame simply by averaging corresponding data values in reference measurement (slow
time, see 2.4) and removing it from every frame in the measurement data. This method is widely used in UWB radar background
processing [2,31]. The experimental results in this thesis show that
this technique effectively removes the clutters.
• Removal of absolute maximum: In this method the absolute maximum of corresponding data points of all data in the reference measurement (slow time) is founded. This method can help removing
high peaks showing up in some frames due to the noise. The disadvantage of this method is that it might remove some important
data with high amplitude. This method was developed based on
empirical trials and showed a good performance when an activity
or movement was to be detected in a cluttered environment.
• Median removal: In this method the median of corresponding data
points of all data in the reference measurement (slow time) is removed from the measurement frames. This method was developed
based on empirical trials and showed a good performance when an
activity or movement was to be detected in a cluttered environment.
• Exponential averaging: Exponential averaging algorithm, computes
a new background estimate yk from the previous background yk−1
that is updated by a new measured radar return xk
yk = αyk−1 + (1 − α)xk (3.2)
where α is the constant scalar weighing factor with the value between 0 and 1, yk and xk are vectors with the size of [1× m]
45
34 Chapter 3. System Design and Validation
containing background estimate and measured radar return in fast
time (see 2.4), respectively. Index k refers to the actual time instant. Thus, new background estimate takes a fraction of the previous estimate and a fraction from the measured radar response.
By controlling the α value it is possible to emphasize the recent
events, or smoothing out high frequency variations and revealing
long term trends in the background estimation [6].
• Adaptive background subtraction: This method is implemented
based on the method presented by Zetik et al. [6]. This method
has the advantage of removing the background from measurements
where human detection is based on respiration or smaller movements when other background removal methods remove these small
movement as they are static backgrounds. In this method α is not
a scalar value but a vector of weighing coefficients αk. This vector
is time variant where k is a certain time instance. Vector qk is
adapting the αk coefficients. The following lines shows how α is
changed in an adaptive way.
if qik < T hreshold1
if qik/zik < T hreshold2
αik = 1;
else
αik = α;
else
αik = 1;
This method is implemented in this thesis and showed a good performance in removing the background for the walking human detection scenario.
3.3.3 Target detection
Detection algorithms will determine if the target is absent or present in
radar signal.
• CFAR: CFAR detectors provide adaptive estimation of an optimum
threshold based on Neyman-Pearson criterion and assuming that
the probability distribution function of the clutter is known [36].
CFAR detection is widely used in UWB radar target detection
[4, 8, 19]. An overview depiction of CFAR is presented in Figure
3.9. CFAR is an adaptive procedure that detects the targets by
34 Chapter 3. System Design and Validation
containing background estimate and measured radar return in fast
time (see 2.4), respectively. Index k refers to the actual time instant. Thus, new background estimate takes a fraction of the previous estimate and a fraction from the measured radar response.
By controlling the α value it is possible to emphasize the recent
events, or smoothing out high frequency variations and revealing
long term trends in the background estimation [6].
• Adaptive background subtraction: This method is implemented
based on the method presented by Zetik et al. [6]. This method
has the advantage of removing the background from measurements
where human detection is based on respiration or smaller movements when other background removal methods remove these small
movement as they are static backgrounds. In this method α is not
a scalar value but a vector of weighing coefficients αk. This vector
is time variant where k is a certain time instance. Vector qk is
adapting the αk coefficients. The following lines shows how α is
changed in an adaptive way.
if qik < T hreshold1
if qik/zik < T hreshold2
αik = 1;
else
αik = α;
else
αik = 1;
This method is implemented in this thesis and showed a good performance in removing the background for the walking human detection scenario.
3.3.3 Target detection
Detection algorithms will determine if the target is absent or present in
radar signal.
• CFAR: CFAR detectors provide adaptive estimation of an optimum
threshold based on Neyman-Pearson criterion and assuming that
the probability distribution function of the clutter is known [36].
CFAR detection is widely used in UWB radar target detection
[4, 8, 19]. An overview depiction of CFAR is presented in Figure
3.9. CFAR is an adaptive procedure that detects the targets by
46
3.3 Signal Processing Algorithms 35
Figure 3.9: CFAR processing.
comparing the neighboring cells to the cell under test and a decision
is made about if a cell under test is a target or not. The advantage
of this method is that the adaptive threshold keeps the probability
of false positive at a constant value. In this thesis, CFAR detector
is used for human walking detection and the results are presented
in paper (B).
• Peak detector: This method detects the peaks in the range measurement data (fast time) based on a certain limit. These peaks
are presented as potential targets. This method is used in this
thesis for human detection and tracking and provided a good performance.
• Integral detector: Sometimes the radar return of an object does not
produce a high enough peak to be detected by the peak detector.
The integral detector calculates the integral under the peaks and
present them as a target if the integral value is greater than a certain limit. This method is used in this thesis for human detection
and tracking and provided a good performance.
3.3.4 Target Tracking
The target tracker algorithm developed in this thesis is designed based
on the combination of the peak detector and the integral detector. A
3.3 Signal Processing Algorithms 35
Figure 3.9: CFAR processing.
comparing the neighboring cells to the cell under test and a decision
is made about if a cell under test is a target or not. The advantage
of this method is that the adaptive threshold keeps the probability
of false positive at a constant value. In this thesis, CFAR detector
is used for human walking detection and the results are presented
in paper (B).
• Peak detector: This method detects the peaks in the range measurement data (fast time) based on a certain limit. These peaks
are presented as potential targets. This method is used in this
thesis for human detection and tracking and provided a good performance.
• Integral detector: Sometimes the radar return of an object does not
produce a high enough peak to be detected by the peak detector.
The integral detector calculates the integral under the peaks and
present them as a target if the integral value is greater than a certain limit. This method is used in this thesis for human detection
and tracking and provided a good performance.
3.3.4 Target Tracking
The target tracker algorithm developed in this thesis is designed based
on the combination of the peak detector and the integral detector. A
47
36 Chapter 3. System Design and Validation
Figure 3.10: Flow chart of the target tracker algorithm.
flow chart of this algorithm is shown in 3.10. The target tracker algorithms is used for detection of walking human. In paper (B) the target
tracker algorithms is used in combination with CFAR and provided good
results in an anechoic chamber. The algorithm does not provide a good
performance in the office case (the rate of false positive is quite high
(51%)).
3.3.5 Other Algorithms
In this thesis, other algorithms are developed for other applications than
target detection and tracking. Two of these algorithms are presented
below.
Nonlinear distance mapping
Nonlinear distance mapping maps the data to get its amplitudes to be
independent of the distance i.e. echos from far shall be able to ”compete”
with echos from near. This is mostly due to a better representation of
the targets at larger distances. It has been observed that the modulation
is not linear by measurements of a metal plate at different distances. By
using the fit function from MATLAB the following polynomial function
was estimated:
f(x) = 1461x
2 − 9119x + 15687 (3.3)
where x is every measurement point of the radar data.
36 Chapter 3. System Design and Validation
Figure 3.10: Flow chart of the target tracker algorithm.
flow chart of this algorithm is shown in 3.10. The target tracker algorithms is used for detection of walking human. In paper (B) the target
tracker algorithms is used in combination with CFAR and provided good
results in an anechoic chamber. The algorithm does not provide a good
performance in the office case (the rate of false positive is quite high
(51%)).
3.3.5 Other Algorithms
In this thesis, other algorithms are developed for other applications than
target detection and tracking. Two of these algorithms are presented
below.
Nonlinear distance mapping
Nonlinear distance mapping maps the data to get its amplitudes to be
independent of the distance i.e. echos from far shall be able to ”compete”
with echos from near. This is mostly due to a better representation of
the targets at larger distances. It has been observed that the modulation
is not linear by measurements of a metal plate at different distances. By
using the fit function from MATLAB the following polynomial function
was estimated:
f(x) = 1461x
2 − 9119x + 15687 (3.3)
where x is every measurement point of the radar data.
48
3.3 Signal Processing Algorithms 37
Up-sampling
Due to the stability of the signal in x-axis by up-sampling the signal
it is possible to reach a better accuracy. Theoretically the radar range
resolution is 7.5 cm based on the center frequency. By up-sampling
the signal 20 times it was possible to reach around 3 mm precision in
range resolution. The increased resolution made it possible to detect
fine movements such as movement of the metal plate in simulation of
the respiratory activity in 3.2.4 .
3.3 Signal Processing Algorithms 37
Up-sampling
Due to the stability of the signal in x-axis by up-sampling the signal
it is possible to reach a better accuracy. Theoretically the radar range
resolution is 7.5 cm based on the center frequency. By up-sampling
the signal 20 times it was possible to reach around 3 mm precision in
range resolution. The increased resolution made it possible to detect
fine movements such as movement of the metal plate in simulation of
the respiratory activity in 3.2.4 .
49
Chapter 4. Contribution
Chapter 4
Contribution
In this chapter a summary of the published papers A-C, the scientific
contribution of the papers and the author’s contributions are presented.
• Paper A Experimental Comparison Study of UWB Technologies
for Static Human Detection
Melika Hozhabri, Magnus Otterskog, Nikola Petrovic and Martin
Ekstr¨om
IEEE International Conference on Ubiquitous Wireless Broadband
(ICUWB 2016), Nanjing, China.
Abstract: This paper compares two dominant Ultra Wide-Band
(UWB) radar technologies Impulse and M-sequence for static human being detection in free space. The hardware and software
platform for each system is described separately. These two radar
platform performances are tested in real conditions and the results
show that M-sequence UWB radar is better suited for detecting
the static human target in larger distances.
Contribution: A quantitative measurement campaign for comparison of two most used radar technologies for static human detection
is done. These measurements can be used as an indication of performances of these two technologies for human being detection.
Author’s contribution: I have been the main author of this paper
and a major part of the idea was mine. I have planned and performed the measurements with help of Radarbolaget that provided
38
Chapter 4. Contribution
Chapter 4
Contribution
In this chapter a summary of the published papers A-C, the scientific
contribution of the papers and the author’s contributions are presented.
• Paper A Experimental Comparison Study of UWB Technologies
for Static Human Detection
Melika Hozhabri, Magnus Otterskog, Nikola Petrovic and Martin
Ekstr¨om
IEEE International Conference on Ubiquitous Wireless Broadband
(ICUWB 2016), Nanjing, China.
Abstract: This paper compares two dominant Ultra Wide-Band
(UWB) radar technologies Impulse and M-sequence for static human being detection in free space. The hardware and software
platform for each system is described separately. These two radar
platform performances are tested in real conditions and the results
show that M-sequence UWB radar is better suited for detecting
the static human target in larger distances.
Contribution: A quantitative measurement campaign for comparison of two most used radar technologies for static human detection
is done. These measurements can be used as an indication of performances of these two technologies for human being detection.
Author’s contribution: I have been the main author of this paper
and a major part of the idea was mine. I have planned and performed the measurements with help of Radarbolaget that provided
38
50
39
the hardware. I wrote most of the paper and analyzed the results.
• Paper B Study of Environment Effect on Detection of Walking
Human by M-Sequence UWB Radar
Melika Hozhabri, Per-Olov Risman and Nikola Petrovic 2016 IEEE
Conference on Antenna Measurements & Applications (CAMA)
Abstract : This paper presents an experimental comparison study
of human movement and presence detection in different environments using ultra wide-band (UWB) M-Sequence radar. The benchmarking measurements are made in an anechoic chamber and repeated in an open office environment. The wave forms of the background noise and scattered amplitudes of a human body are measured and compared. A set of detection algorithms and filters
which are developed to track the human movement and presence
is presented and the tracking results in these two environments are
compared to each other.
Contribution: These quantitative measurements show that the
multipath reflection has most contribution of false alarm. The
target tracking algorithm is developed from scratch based on some
input argument from the user.
Author’s contribution: I have been the main author of this paper. I
have planned and performed the measurements with help of Delta
that provided the semi-anechoic chamber. I wrote most of the
paper and partly analyzed the results. I partly developed the signal
processing algorithms and filters.
• Paper C Comparison of UWB Radar Backscattering by the Human Torso and a Phantom
Melika Hozhabri, Per-Olov Risman and Nikola Petrovic 2018 IEEE
Conference on Antenna Measurements & Applications (CAMA)
Abstract: An Ultra Wide-Band (UWB) radar is used to measure
the backscattering of a human and a human phantom. The choice
of material and shape for the human phantom is discussed. The
dielectric properties of the material (wet sand) used in the experiment are measured by a retromodeling technique and also calculated by mixture formulas. The appropriate frequency choice for
the application is discussed.
Contribution: A phantom of human trunk is developed and tested.
39
the hardware. I wrote most of the paper and analyzed the results.
• Paper B Study of Environment Effect on Detection of Walking
Human by M-Sequence UWB Radar
Melika Hozhabri, Per-Olov Risman and Nikola Petrovic 2016 IEEE
Conference on Antenna Measurements & Applications (CAMA)
Abstract : This paper presents an experimental comparison study
of human movement and presence detection in different environments using ultra wide-band (UWB) M-Sequence radar. The benchmarking measurements are made in an anechoic chamber and repeated in an open office environment. The wave forms of the background noise and scattered amplitudes of a human body are measured and compared. A set of detection algorithms and filters
which are developed to track the human movement and presence
is presented and the tracking results in these two environments are
compared to each other.
Contribution: These quantitative measurements show that the
multipath reflection has most contribution of false alarm. The
target tracking algorithm is developed from scratch based on some
input argument from the user.
Author’s contribution: I have been the main author of this paper. I
have planned and performed the measurements with help of Delta
that provided the semi-anechoic chamber. I wrote most of the
paper and partly analyzed the results. I partly developed the signal
processing algorithms and filters.
• Paper C Comparison of UWB Radar Backscattering by the Human Torso and a Phantom
Melika Hozhabri, Per-Olov Risman and Nikola Petrovic 2018 IEEE
Conference on Antenna Measurements & Applications (CAMA)
Abstract: An Ultra Wide-Band (UWB) radar is used to measure
the backscattering of a human and a human phantom. The choice
of material and shape for the human phantom is discussed. The
dielectric properties of the material (wet sand) used in the experiment are measured by a retromodeling technique and also calculated by mixture formulas. The appropriate frequency choice for
the application is discussed.
Contribution: A phantom of human trunk is developed and tested.
51
40 Chapter 4. Contribution
The quantitative measurements show that the phantom can represent the human trunk in radar measurements. It is also shown
that UWB radar has the capability of detect the breast/neck/head
distance differences.
Author’s contribution: I have been the main author of this paper. I
have planned and performed the measurements with help of Radarbolaget in G¨avle University that provided the radar hardware. I
wrote most of the paper and partly analyzed the results.
40 Chapter 4. Contribution
The quantitative measurements show that the phantom can represent the human trunk in radar measurements. It is also shown
that UWB radar has the capability of detect the breast/neck/head
distance differences.
Author’s contribution: I have been the main author of this paper. I
have planned and performed the measurements with help of Radarbolaget in G¨avle University that provided the radar hardware. I
wrote most of the paper and partly analyzed the results.
52
Chapter 5
Conclusion and Future
Work
This chapter presents the summary of this thesis and the possible future
directions.
5.1 Conclusion
This thesis focuses on human detection in enclosed industrial environments with UWB radar and is performed in collaboration with industry.
It therefore deals with a real world application where the requirements
are well defined and rigid, and false positive and negative outputs can be
costly in robotic or a mining applications as well as in terms of injuries
and even loss of life.
M-sequence and pulse radar are two commercial UWB radar systems
that are widely used for human detection in literature but their performance were not compared in an experimental set-up. An experimental
measurement of a static human being at different distances is performed
in paper (A). The results show that M-sequence UWB radar is better
suited for detecting the static human target in larger distances at the
price of relative higher power consumption.
Humans have different body structure and clothing and that affect
the measurements. To make the measurements more reliable for different
set-ups a phantom of human torso is built based on the materials that
41
Chapter 5
Conclusion and Future
Work
This chapter presents the summary of this thesis and the possible future
directions.
5.1 Conclusion
This thesis focuses on human detection in enclosed industrial environments with UWB radar and is performed in collaboration with industry.
It therefore deals with a real world application where the requirements
are well defined and rigid, and false positive and negative outputs can be
costly in robotic or a mining applications as well as in terms of injuries
and even loss of life.
M-sequence and pulse radar are two commercial UWB radar systems
that are widely used for human detection in literature but their performance were not compared in an experimental set-up. An experimental
measurement of a static human being at different distances is performed
in paper (A). The results show that M-sequence UWB radar is better
suited for detecting the static human target in larger distances at the
price of relative higher power consumption.
Humans have different body structure and clothing and that affect
the measurements. To make the measurements more reliable for different
set-ups a phantom of human torso is built based on the materials that
41
53
42 Chapter 5. Conclusion and Future Work
have similar electromagnetic properties as human body. The results are
presented in paper (C) and it is answering the research question 3.
The UWB radar system shows some good capabilities for human
detection in short distances and due to its fine resolution in time it is
possible to estimate the target distance to the radar device with good
precision. The signal amplitude is not reliable as it shows relatively big
fluctuations in different scans therefore signal amplitude data is not used
in processing of the data. Signal amplitude can be used as a feature in
signal processing algorithms for better detection of the target.
There are several challenges for reaching the research goal (see 1.2).
It is also important to understand the effect of clutter on signal detection.
Based of the measurements in the office environment and published in
paper (B) the probability of false positive and negative outputs in a
cluttered environment is around 50%. This is not a satisfactory result
and needs further investigation.
In addition, discrimination of a human from other object is not an
easy task. Spatio-temporal properties of a dynamic human body such as
arm and hand movements could be a discerned by micro-Doppler processing which is not fully examined in this thesis but in case of static
humans the chest movement during breathing could not deliver satisfactory results in a real world scenario and needs further investigations.
Furthermore based on the Mie theory presented in paper (C) the 1–3
GHz system does not provide the best discerning for an object with the
size and geometry of the human body and use of lower frequencies is
recommended.
5.2 Future Work
Future work can be directed in different paths and some of them are
presented in this section.
During the work within this thesis, some of the system characteristics
such as precision and effect of environmental noise in the signal detection were measured and defined, whereas some other characteristics such
as the maximum range needs to be further investigated. Furthermore,
additional investigations are needed to determine the origin of the high
percentage of the false positive and negative outputs; how large part
that is dependent on the signal processing algorithms and dependent on
the system/hardware, respectively.
42 Chapter 5. Conclusion and Future Work
have similar electromagnetic properties as human body. The results are
presented in paper (C) and it is answering the research question 3.
The UWB radar system shows some good capabilities for human
detection in short distances and due to its fine resolution in time it is
possible to estimate the target distance to the radar device with good
precision. The signal amplitude is not reliable as it shows relatively big
fluctuations in different scans therefore signal amplitude data is not used
in processing of the data. Signal amplitude can be used as a feature in
signal processing algorithms for better detection of the target.
There are several challenges for reaching the research goal (see 1.2).
It is also important to understand the effect of clutter on signal detection.
Based of the measurements in the office environment and published in
paper (B) the probability of false positive and negative outputs in a
cluttered environment is around 50%. This is not a satisfactory result
and needs further investigation.
In addition, discrimination of a human from other object is not an
easy task. Spatio-temporal properties of a dynamic human body such as
arm and hand movements could be a discerned by micro-Doppler processing which is not fully examined in this thesis but in case of static
humans the chest movement during breathing could not deliver satisfactory results in a real world scenario and needs further investigations.
Furthermore based on the Mie theory presented in paper (C) the 1–3
GHz system does not provide the best discerning for an object with the
size and geometry of the human body and use of lower frequencies is
recommended.
5.2 Future Work
Future work can be directed in different paths and some of them are
presented in this section.
During the work within this thesis, some of the system characteristics
such as precision and effect of environmental noise in the signal detection were measured and defined, whereas some other characteristics such
as the maximum range needs to be further investigated. Furthermore,
additional investigations are needed to determine the origin of the high
percentage of the false positive and negative outputs; how large part
that is dependent on the signal processing algorithms and dependent on
the system/hardware, respectively.
54
5.2 Future Work 43
In paper (B) the choice of the frequency for human body is discussed.
Based on physical dimensions of the human body and the human tissues
permittivity, the discerning of a human body are likely improved by
choosing lower operating frequencies typically around 70 to 200 MHz,
since these could improve the discrimination of the trunk by its size in
relation to the wavelength.
Using the Multiple Input Multiple Output (MIMO) system can reveal
more details about the targets. Due to scalability of the chosen UWB
system it is possible to connect 12 antenna pairs to it and subsequently
gain more information about the target.
One other possibility is to use multiple sensors to take advantage of
each sensor strength to increased confidence of detection and decrease
false positive and negative outputs. Combination of radar sensor with
cameras may be a next logical step.
Last but not least, other signal and data processing approaches such
as Machine Learning (ML) and Artificial Intelligence (AI) can be used
for detection, target tracking and target classification that can improve
the target detection and classification.
5.2 Future Work 43
In paper (B) the choice of the frequency for human body is discussed.
Based on physical dimensions of the human body and the human tissues
permittivity, the discerning of a human body are likely improved by
choosing lower operating frequencies typically around 70 to 200 MHz,
since these could improve the discrimination of the trunk by its size in
relation to the wavelength.
Using the Multiple Input Multiple Output (MIMO) system can reveal
more details about the targets. Due to scalability of the chosen UWB
system it is possible to connect 12 antenna pairs to it and subsequently
gain more information about the target.
One other possibility is to use multiple sensors to take advantage of
each sensor strength to increased confidence of detection and decrease
false positive and negative outputs. Combination of radar sensor with
cameras may be a next logical step.
Last but not least, other signal and data processing approaches such
as Machine Learning (ML) and Artificial Intelligence (AI) can be used
for detection, target tracking and target classification that can improve
the target detection and classification.
55
Chapter A. Abbrevations
Appendix A
Abbrevations
AWGN Additive White Gaussian Noise
AIK Akaike Information Criterion
AUT Antenna Under Test
CFAR Constant False Alarm Rate
CWT Continuous Wavelet Transform
EMI Electromagnetic Interference
EM Electromagnetic
FDTD Finite Difference Time Domain
FFT Fast Fourier Transform
FCC Federal Communications Commission
FMCW Frequency Modulated Continuous Wave
GPS Global Positioning System
HMM Hidden Markov Model
IIR Infinite Impulse Response
LOS Line Of Sight
MA Measurement Antenna
MTI Moving Target Indication
MTT Multi-Target Tracking
PCA Principal Component Analysis
PDF Probability Density Function
44
Chapter A. Abbrevations
Appendix A
Abbrevations
AWGN Additive White Gaussian Noise
AIK Akaike Information Criterion
AUT Antenna Under Test
CFAR Constant False Alarm Rate
CWT Continuous Wavelet Transform
EMI Electromagnetic Interference
EM Electromagnetic
FDTD Finite Difference Time Domain
FFT Fast Fourier Transform
FCC Federal Communications Commission
FMCW Frequency Modulated Continuous Wave
GPS Global Positioning System
HMM Hidden Markov Model
IIR Infinite Impulse Response
LOS Line Of Sight
MA Measurement Antenna
MTI Moving Target Indication
MTT Multi-Target Tracking
PCA Principal Component Analysis
PDF Probability Density Function
44
56
45
RCS Radar Cross Section
RPU Radar Processing Unit
RPS Range Profile Subtraction
RMD Respiratory Motion Detection
RMS Root Mean Square
RQ Research Question
STA/LTA Short Term to Long Term Average
SMPM Specular Multi-Path Model
SVM Support Vector Machine
TOA Time Of Arrival
UWB Ultra Wide-Band
WRT Wide-band Radar Transceiver
45
RCS Radar Cross Section
RPU Radar Processing Unit
RPS Range Profile Subtraction
RMD Respiratory Motion Detection
RMS Root Mean Square
RQ Research Question
STA/LTA Short Term to Long Term Average
SMPM Specular Multi-Path Model
SVM Support Vector Machine
TOA Time Of Arrival
UWB Ultra Wide-Band
WRT Wide-band Radar Transceiver
57
58
Bibliography
[1] S. Chang, M. Wolf, and J. W. Burdick. Human detection and tracking via ultra-wideband (uwb) radar. In 2010 IEEE International
Conference on Robotics and Automation, pages 452–457, May 2010.
[2] Y Kilic, H Wymeersch, A Meijerink, M J Bentum, and W G Scanlon. An experimental study of UWB device-free person detection
and ranging. In Ultra-Wideband (ICUWB), 2013 IEEE International Conference on, pages 43–48, sep 2013.
[3] Go Shingu, Kenichi Takizawa, and Tetushi Ikegami. Human body
detection using UWB radar in an indoor environment. 2008 International Symposium on Communications and Information Technologies, ISCIT 2008, 77046:283–288, 2008.
[4] J. Sachs, M. Aftanas, S. Crabbe, M. Drutarovsky, R. Klukas,
D. Kocur, T. T. Nguyen, P. Peyerl, J. Rovnakova, and E. Zaikov.
Detection and tracking of moving or trapped people hidden by obstacles using ultra-wideband pseudo-noise radar. In 2008 European
Radar Conference, pages 408–411, Oct 2008.
[5] S. A. Rane, S. Sarkar, and A. Gaurav. Moving target localization
using ultra wideband sensing. In 2016 Online International Conference on Green Engineering and Technologies (IC-GET), pages 1–5,
Nov 2016.
[6] Rudolf Zetik, Stephen Crabbe, Jozef Krajnak, Peter Peyerl, J¨urgen
Sachs, and Reiner Thom¨a. Detection and localization of persons
behind obstacles using M-sequence through-the-wall radar, 2006.
47
Bibliography
[1] S. Chang, M. Wolf, and J. W. Burdick. Human detection and tracking via ultra-wideband (uwb) radar. In 2010 IEEE International
Conference on Robotics and Automation, pages 452–457, May 2010.
[2] Y Kilic, H Wymeersch, A Meijerink, M J Bentum, and W G Scanlon. An experimental study of UWB device-free person detection
and ranging. In Ultra-Wideband (ICUWB), 2013 IEEE International Conference on, pages 43–48, sep 2013.
[3] Go Shingu, Kenichi Takizawa, and Tetushi Ikegami. Human body
detection using UWB radar in an indoor environment. 2008 International Symposium on Communications and Information Technologies, ISCIT 2008, 77046:283–288, 2008.
[4] J. Sachs, M. Aftanas, S. Crabbe, M. Drutarovsky, R. Klukas,
D. Kocur, T. T. Nguyen, P. Peyerl, J. Rovnakova, and E. Zaikov.
Detection and tracking of moving or trapped people hidden by obstacles using ultra-wideband pseudo-noise radar. In 2008 European
Radar Conference, pages 408–411, Oct 2008.
[5] S. A. Rane, S. Sarkar, and A. Gaurav. Moving target localization
using ultra wideband sensing. In 2016 Online International Conference on Green Engineering and Technologies (IC-GET), pages 1–5,
Nov 2016.
[6] Rudolf Zetik, Stephen Crabbe, Jozef Krajnak, Peter Peyerl, J¨urgen
Sachs, and Reiner Thom¨a. Detection and localization of persons
behind obstacles using M-sequence through-the-wall radar, 2006.
47
59
48 Bibliography
[7] Amer Nezirovi´c, Alexander G. Yarovoy, and Leo P. Ligthart. Signal
processing for improved detection of trapped victims using UWB
radar. IEEE Transactions on Geoscience and Remote Sensing, 48(4
PART 2):2005–2014, apr 2010.
[8] Jana Rov´akov´a, Duˇsan Kocur, and Peter Kaˇzim´ır. Investigation of
localization accuracy for UWB radar operating in complex environment. Acta Polytechnica Hungarica, 10(5):203–219, 2013.
[9] G. Zhao, Q. Liang, and T. S. Durrani. Uwb radar target detection
based on hidden markov models. IEEE Access, 6:28702–28711, 2018.
[10] S. M. Patole, M. Torlak, D. Wang, and M. Ali. Automotive radars:
A review of signal processing techniques. IEEE Signal Processing
Magazine, 34(2):22–35, March 2017.
[11] M. Zlatanski, P. Sommer, F. Zurfluh, and G. L. Madonna. Radar
sensor for fenceless machine guarding and collaborative robotics. In
2018 IEEE International Conference on Intelligence and Safety for
Robotics (ISR), pages 19–25, Aug 2018.
[12] Thiago Teixeira, Gershon Dublon, and Andreas Savvides. A Survey
of Human-Sensing: Methods for Detecting Presence, Count, Location, Track, and Identity. ACM Computing Surveys, 5:1–35, 2010.
[13] Yakup Kilic, Henk Wymeersch, Arjan Meijerink, Mark J. Bentum,
and William G. Scanlon. Device-free person detection and ranging in UWB networks. IEEE Journal on Selected Topics in Signal
Processing, 8(1):43–54, 2014.
[14] Sang Hyun Chang, Naoki Mitsumoto, and Joel W. Burdick. An
algorithm for UWB radar-based human detection. IEEE National
Radar Conference - Proceedings, pages 0–5, 2009.
[15] M Imran and A Shoaib. Detection of multiple targets using ultrawideband radar. Master thesis, G¨avle University, Sweden, Feb.2011.
[16] H. G. Schantz. Introduction to ultra-wideband antennas. In IEEE
Conference on Ultra Wideband Systems and Technologies, 2003,
pages 1–9, Nov 2003.
48 Bibliography
[7] Amer Nezirovi´c, Alexander G. Yarovoy, and Leo P. Ligthart. Signal
processing for improved detection of trapped victims using UWB
radar. IEEE Transactions on Geoscience and Remote Sensing, 48(4
PART 2):2005–2014, apr 2010.
[8] Jana Rov´akov´a, Duˇsan Kocur, and Peter Kaˇzim´ır. Investigation of
localization accuracy for UWB radar operating in complex environment. Acta Polytechnica Hungarica, 10(5):203–219, 2013.
[9] G. Zhao, Q. Liang, and T. S. Durrani. Uwb radar target detection
based on hidden markov models. IEEE Access, 6:28702–28711, 2018.
[10] S. M. Patole, M. Torlak, D. Wang, and M. Ali. Automotive radars:
A review of signal processing techniques. IEEE Signal Processing
Magazine, 34(2):22–35, March 2017.
[11] M. Zlatanski, P. Sommer, F. Zurfluh, and G. L. Madonna. Radar
sensor for fenceless machine guarding and collaborative robotics. In
2018 IEEE International Conference on Intelligence and Safety for
Robotics (ISR), pages 19–25, Aug 2018.
[12] Thiago Teixeira, Gershon Dublon, and Andreas Savvides. A Survey
of Human-Sensing: Methods for Detecting Presence, Count, Location, Track, and Identity. ACM Computing Surveys, 5:1–35, 2010.
[13] Yakup Kilic, Henk Wymeersch, Arjan Meijerink, Mark J. Bentum,
and William G. Scanlon. Device-free person detection and ranging in UWB networks. IEEE Journal on Selected Topics in Signal
Processing, 8(1):43–54, 2014.
[14] Sang Hyun Chang, Naoki Mitsumoto, and Joel W. Burdick. An
algorithm for UWB radar-based human detection. IEEE National
Radar Conference - Proceedings, pages 0–5, 2009.
[15] M Imran and A Shoaib. Detection of multiple targets using ultrawideband radar. Master thesis, G¨avle University, Sweden, Feb.2011.
[16] H. G. Schantz. Introduction to ultra-wideband antennas. In IEEE
Conference on Ultra Wideband Systems and Technologies, 2003,
pages 1–9, Nov 2003.
60
Bibliography 49
[17] W. Wiesbeck, G. Adamiuk, and C. Sturm. Basic properties and design principles of uwb antennas. Proceedings of the IEEE, 97(2):372–
385, Feb 2009.
[18] Otar Javashvili. UWB-Antennas. Master’s thesis, University of
G¨avle, 2009.
[19] Jana Rov´akov´a and Duˇsan Kocur. UWB Radar Signal Processing
for Positioning of Persons Changing Their Motion Activity. Acta
Polytechnica Hungarica, 10:165–184, 2013.
[20] D. A. Shnidman. Generalized radar clutter model. IEEE Transactions on Aerospace and Electronic Systems, 35(3):857–865, Jul
1999.
[21] J. D. Bryan, J. Kwon, N. Lee, and Y. Kim. Application of ultrawide band radar for classification of human activities. IET Radar,
Sonar Navigation, 6(3):172–179, March 2012.
[22] Yazhou Wang. UWB Pulse Radar for Human Imaging and Doppler
Detection Applications. PhD dissertation, University of Tennessee,
may 2012.
[23] V. C. Chen. Detection and analysis of human motion by radar. In
2008 IEEE Radar Conference, pages 1–4, May 2008.
[24] James D. Taylor. Ultrawideband Radar Technology. CRC press,
2000.
[25] Jeffrey A. Nanzer. A Review of Microwave Wireless Techniques for
Human Presence Detection and Classification. IEEE Transactions
on Microwave Theory and Techniques, 65(5):1780–1794, 2017.
[26] B. Denis, L. Ouvry, B. Uguen, and F. Tehoffo-Talom. Advanced
Bayesian Filtering Techniques for UWB Tracking Systems in Indoor
Environments. In 2005 IEEE International Conference on UltraWideband, pages 638–643. IEEE.
[27] Bita Sobhani, Enrico Paolini, Andrea Giorgetti, Matteo Mazzotti,
and Marco Chiani. Target Tracking for UWB Multistatic Radar
Sensor Networks. IEEE Journal of Selected Topics in Signal Processing, 8(1):125–136, feb 2014.
Bibliography 49
[17] W. Wiesbeck, G. Adamiuk, and C. Sturm. Basic properties and design principles of uwb antennas. Proceedings of the IEEE, 97(2):372–
385, Feb 2009.
[18] Otar Javashvili. UWB-Antennas. Master’s thesis, University of
G¨avle, 2009.
[19] Jana Rov´akov´a and Duˇsan Kocur. UWB Radar Signal Processing
for Positioning of Persons Changing Their Motion Activity. Acta
Polytechnica Hungarica, 10:165–184, 2013.
[20] D. A. Shnidman. Generalized radar clutter model. IEEE Transactions on Aerospace and Electronic Systems, 35(3):857–865, Jul
1999.
[21] J. D. Bryan, J. Kwon, N. Lee, and Y. Kim. Application of ultrawide band radar for classification of human activities. IET Radar,
Sonar Navigation, 6(3):172–179, March 2012.
[22] Yazhou Wang. UWB Pulse Radar for Human Imaging and Doppler
Detection Applications. PhD dissertation, University of Tennessee,
may 2012.
[23] V. C. Chen. Detection and analysis of human motion by radar. In
2008 IEEE Radar Conference, pages 1–4, May 2008.
[24] James D. Taylor. Ultrawideband Radar Technology. CRC press,
2000.
[25] Jeffrey A. Nanzer. A Review of Microwave Wireless Techniques for
Human Presence Detection and Classification. IEEE Transactions
on Microwave Theory and Techniques, 65(5):1780–1794, 2017.
[26] B. Denis, L. Ouvry, B. Uguen, and F. Tehoffo-Talom. Advanced
Bayesian Filtering Techniques for UWB Tracking Systems in Indoor
Environments. In 2005 IEEE International Conference on UltraWideband, pages 638–643. IEEE.
[27] Bita Sobhani, Enrico Paolini, Andrea Giorgetti, Matteo Mazzotti,
and Marco Chiani. Target Tracking for UWB Multistatic Radar
Sensor Networks. IEEE Journal of Selected Topics in Signal Processing, 8(1):125–136, feb 2014.
61
[28] M.I. Skolnik. Radar Handbook, Third Edition. Electronics electrical
engineering. McGraw-Hill Education, 2008.
[29] Traian Dogaru, Lam Nguyen, and Calvin Le. Computer models
of the human body signature for sensing through the wall radar
applications. Sensors (Peterborough, NH), (September):56, 2007.
[30] Naoyuki Yamada, Yuichi Tanaka, and Kunitoshi Nishikawa. Radar
cross section for pedestrian in 76GHz band. In 35th European Microwave Conference 2005 - Conference Proceedings, volume 2, pages
1015–1018, 2005.
[31] G. Ossberger, T. Buchegger, E. Schimback, A. Stelzer, and
R. Weigel. Non-invasive respiratory movement detection and monitoring of hidden humans using ultra wideband pulse radar. In
2004 International Workshop on Ultra Wideband Systems Joint with
Conference on Ultra Wideband Systems and Technologies. Joint
UWBST IWUWBS 2004 (IEEE Cat. No.04EX812), pages 395–399,
May 2004.
[32] J M Cramer, R A Scholtz, and M Z Win. Evaluation of an UltraWideband Propagation Channel. pages 1–9.
[33] S. S. Ram, C. Christianson, Y. Kim, and H. Ling. Simulation and
analysis of human micro-dopplers in through-wall environments.
IEEE Transactions on Geoscience and Remote Sensing, 48(4):2015–
2023, April 2010.
[34] Michael D. Foegelle. Antenna Pattern Measurement: Concepts and
Techniques. Technical report.
[35] M Kmec1 P Peyerl2 J. Sachs1 R. Herrmann1. Stimulation of UWBsensors: pulse or maximum sequence?
[36] D. Urdzk and D. Kocur. Cfar detectors for through wall tracking
of moving targets by m-sequence uwb radar. In 20th International
Conference Radioelektronika 2010, pages 1–4, April 2010.
[28] M.I. Skolnik. Radar Handbook, Third Edition. Electronics electrical
engineering. McGraw-Hill Education, 2008.
[29] Traian Dogaru, Lam Nguyen, and Calvin Le. Computer models
of the human body signature for sensing through the wall radar
applications. Sensors (Peterborough, NH), (September):56, 2007.
[30] Naoyuki Yamada, Yuichi Tanaka, and Kunitoshi Nishikawa. Radar
cross section for pedestrian in 76GHz band. In 35th European Microwave Conference 2005 - Conference Proceedings, volume 2, pages
1015–1018, 2005.
[31] G. Ossberger, T. Buchegger, E. Schimback, A. Stelzer, and
R. Weigel. Non-invasive respiratory movement detection and monitoring of hidden humans using ultra wideband pulse radar. In
2004 International Workshop on Ultra Wideband Systems Joint with
Conference on Ultra Wideband Systems and Technologies. Joint
UWBST IWUWBS 2004 (IEEE Cat. No.04EX812), pages 395–399,
May 2004.
[32] J M Cramer, R A Scholtz, and M Z Win. Evaluation of an UltraWideband Propagation Channel. pages 1–9.
[33] S. S. Ram, C. Christianson, Y. Kim, and H. Ling. Simulation and
analysis of human micro-dopplers in through-wall environments.
IEEE Transactions on Geoscience and Remote Sensing, 48(4):2015–
2023, April 2010.
[34] Michael D. Foegelle. Antenna Pattern Measurement: Concepts and
Techniques. Technical report.
[35] M Kmec1 P Peyerl2 J. Sachs1 R. Herrmann1. Stimulation of UWBsensors: pulse or maximum sequence?
[36] D. Urdzk and D. Kocur. Cfar detectors for through wall tracking
of moving targets by m-sequence uwb radar. In 20th International
Conference Radioelektronika 2010, pages 1–4, April 2010.
62
