BACHELOR'S THESIS
Naturalistic Observations of Driver and VRU Interactions
Daniel Castro Larsson 2016
Bachelor of Science in Engineering Technology Automotive Engineering
Luleå University of Technology
Department of Computer Science, Electrical and Space Engineering
Abstract
Due to the high number road traffic related fatalities of VRU1 in the EU, it is necessary to improve the performance of the AEB2 systems. This thesis covers the pre-stage of the AEB-sensor development, from the pre-study of different European accident databases to get a better understanding of which type of accidents are the most common and which cases are the most severe, to the Naturalistic Driving Study performed in Barcelona, to record a big amount of critical situations between the driver and the VRU.
The Naturalistic Driving Study was performed with a vehicle equipped with a Lidar3, two cameras, CAN4-system information and a VBOX. All the critical situations was then analyzed using 58 different parameters, from VRU head and torso orientation, weather conditions to intersection type and relative velocity.
The main objective of this thesis is to present a conclusion of how the VRU and the driver react to a critical situation. This data will be useful for future sensor development for more precise Autonomous Emergency Brake systems and the development of Automatic Emergency Steering.
1Vulnerable Road Users
2Autonomous Emergency Brake
3Light Detection And Ranging
4Controller Area Network
i
ii ABSTRACT
Sammanfattning
P˚a grund av det h¨oga antalet avlidna utsatta v¨ag anv¨andare i trafiken i EU, s˚a beh¨ovs prestandan p˚a dagens autobromssystem f¨orb¨attras. Denna avhandling t¨acker f¨orarbetet som kr¨avs f¨or sensorutveckling av autobromssys- tem. Fr˚an f¨orstudie av olika europeiska olycksdatabaser f¨or att f˚a en b¨attre f¨orst˚aelse av vilken typ av olyckor som ¨ar de vanligaste och vilka som ¨ar de allvarligaste. Till en naturalistisk k¨or studie som utf¨ors i Barcelona, f¨or att samla in en stor m¨angd data fr˚an kritiska situationer mellan f¨oraren och utsatta v¨ag anv¨andare.
Den naturalistiska k¨or studien utf¨ordes med ett fordon utrustat med en Li- dar, tv˚a kameror, CAN5 systeminformation och en VBOX. Alla kritiska sit- uationer analyserades med hj¨alp av 58 olika parametrar, fr˚an utsatta v¨ag anv¨andares huvud och kropp orientering, v¨aderf¨orh˚allanden till vilken typ av infrastruktur och relativ hastighet.
Huvudsyftet med denna avhandling ¨ar att presentera en slutsats om hur den utsatta v¨ag anv¨andaren och f¨oraren reagerar p˚a en kritisk situation. Datan som framst¨alls kommer att vara anv¨andbar f¨or framtida utveckling av sen- sorer f¨or mer precisa autobromssystem och utveckling av mer komplexa sys- tem s˚a som automatisk n¨odstyrning.
5Controller Area Network
iii
iv SAMMANFATTNING
Acknowledgements
It would have not been possible to perform this thesis without the ADAS6 team at Applus IDIADA leaded by Andr´es Aparicio and specially not with- out my supervisor Laura Sanz.
The knowledge achieved from the Automotive Engineering program at Lule˚a University of Technology, leaded by Peter Jeppsson and Kim Berglund, has been crucial to perform this thesis with quality.
And not to forget, the help from my examiner at Lule˚a University of Tech- nology, Jan van Deventer.
6Advanced Driving Assistance Systems
v
vi ACKNOWLEDGEMENTS
Contents
Abstract i
Sammanfattning iii
Acknowledgements v
Glossary xiii
1 Introduction 1
1.1 Background . . . 1
1.1.1 PROSPECT . . . 2
1.1.2 Applus IDIADA . . . 4
1.2 Future benefits . . . 5
1.3 Project boundaries . . . 5
2 Theory 7 2.1 Definitions . . . 7
2.1.1 Time to Collision . . . 7
2.1.2 Post Encroachment Time. . . 8
2.1.3 Forward Collision Warning . . . 9
2.1.4 Human Machine Interface . . . 9
2.1.5 Vulnerable Road Users . . . 9
2.1.6 Absolute velocity . . . 9
2.1.7 Relative velocity . . . 9
2.1.8 Yaw, pitch and roll . . . 9
2.2 Autonomous Emergency Braking . . . 10
2.3 Test equipment . . . 11
2.4 Field Operational Test . . . 13
2.5 DOCTOR-method . . . 14
2.6 Analysis method . . . 15
vii
viii CONTENTS
3 Preparation and Execution of Analysis 17
3.1 Vehicle preparation . . . 17
3.1.1 Mounting of the test equipment . . . 18
3.1.2 Lidar configuration . . . 22
3.1.3 Validation . . . 23
3.2 Designing of the analysis platform . . . 24
3.3 Analysis procedure . . . 25
4 Results 27 4.1 Vehicle equipment . . . 27
4.2 Driver interface in the vehicle . . . 28
4.3 Analysis platform . . . 29
4.4 Analysis results and statistics . . . 30
4.4.1 Global statistics . . . 30
4.4.2 Encounter statistics . . . 32
4.4.3 Intent statistics . . . 35
4.4.4 Kinematic statistics . . . 39
5 Discussion 45
6 Conclusion 47
7 Further work 49
A Project Plan 51
B Parameters 57
C Driver instructions 63
D Gantt 69
Bibliography 71
List of Figures
1.1 The PROSPECT Logo. . . 3
1.2 AEB system under test at Applus IDIADA. . . 4
2.1 Illustration of TTC between two objects. . . 8
2.2 Illustration of PET between two objects. . . 8
2.3 Yaw, pitch and roll illustration. . . 10
2.4 AEB illustration. . . 11
2.5 Layout of the equipment in the vehicle . . . 13
2.6 Recommended NDS structure. . . 14
3.1 The vehicle used in Barcelona, a Peugeot 3008. . . 18
3.2 The Lidar. . . 18
3.3 Ethernet switch and synchronization unit. . . 19
3.4 The camera pointing to the road. . . 19
3.5 The VectorCAN to the left, USB-hub and the trigger to the right. . . 20
3.6 VBOX.. . . 20
3.7 Keypad for the trigger. . . 21
3.8 Fuse box with a switch, hard drive mounted on top. . . 21
3.9 Cooling system. . . 22
3.10 Lidar configuration setup. . . 23
3.11 Layout for controlling. . . 24
3.12 The analysis software vADASdeveloper.. . . 25
4.1 Final equipment layout with the new In-vehicle PC. . . 27
4.2 Overview of the drivers interface. . . 28
4.3 Eco off sticker. . . 29
4.4 Pie chart of the type of VRU. . . 31
4.5 Column plot of the right of way. . . 32
4.6 Column plot of the type of encounter recording to the GIDAS pictogram’s. . . 33
ix
x LIST OF FIGURES
4.7 Radar plot of the VRU head orientation respect the vehicle. . 35
4.8 Figure used during the analysis of the head orientation. . . 36
4.9 Radar plot of the VRU torso orientation respect the vehicle. . 37
4.10 Figure used during the analysis of the torso orientation. . . 38
4.11 Box plot of the vehicle speed. . . 39
4.12 Distribution of the TTC and PET. . . 40
4.13 Plot of TTC. . . 41
4.14 Plot of PET. . . 42
List of Tables
4.1 VRU type distribution . . . 31
4.2 Right of way . . . 32
4.3 Type of encounter. . . 34
4.4 Head orientation distribution . . . 36
4.5 Torso orientation distribution . . . 38
4.6 Statistical values of the vehicle speed . . . 39
4.7 TTC distribution . . . 42
4.8 PET distribution . . . 43
xi
xii LIST OF TABLES
Glossary
The abbreviations we will find in this report are hereinafter explained.
ADAS Advanced Driver Assistance System
AEB Autonomous Emergency Brake
AES Automatic Emergency Steering
CAN Controller Area Network
DOCTOR Dutch Objective Conflict Technique for Operation and Research
ECU Electronic Control Unit
eSATA external Serial AT Attachment
EU European Union
Euro NCAP European New Car Assessment Programme
FCW Forward Collision Warning
FESTA Field Operational Test Support Action
FOT Field Operational Test
GIDAS German In-Depth Accident Study
GPS Global Positioning System
HMI Human Machine Interface
IFSTTAR Institut Fran¸cais des Sciences et Technologies des Transports, de L’am´enagement et des R´eseaux
LIDAR Light Detection And Ranging
LTU Lule˚a University of Technology
MATLAB Matrix Laboratory
NAS Network-attached storage
NDD Naturalistic Driving Data
NDS Naturalistic Driving Study
PC Personal Computer
PET Post Encroachment Time
RADAR Radio Detection And Ranging
xiii
xiv GLOSSARY SQL Structured Query Language
TTC Time To Collision USB Universal Serial Bus VGA Video Graphics Array VRU Vulnerable Road User
Chapter 1 Introduction
This thesis has been carried out with the collaboration of Applus IDIADA and Lule˚a University of Technology as part of the European project, Proac- tive Safety for Pedestrians and Cyclists, PROSPECT.
1.1 Background
Pedestrians, cyclist and moped riders are most vulnerable in road traffic, and are therefore referred to as Vulnerable Road Users (VRU). According to the Community Road Accident Database CARE, the annual number of VRU deaths in the European Union (EU) is 8922. That is 31% of all the deceased road users [1]. The European Union wants to confront this health problem and therefore initiated the PROSPECT project, which will mainly concentrate on cyclists and pedestrians. PROSPECT’s main objective is to get a better understanding of VRU related accidents and the development of new active safety systems, such as Autonomous Emergency Braking sys- tems, AEB. This project includes several partners; Applus IDIADA is one of them. IDIADA is responsible of the coordination of PROSPECT project.
They are also responsible of the naturalistic observations of drivers and VRU interactions in the city of Barcelona. It is in this context that the herewith thesis comes in: it covers all from the preparation of the analysis method, to the preparation of the test vehicles. The main objective of the thesis is to provide a summary and conclusion of the analysed data collected from the observations.
1
2 CHAPTER 1. INTRODUCTION
1.1.1 PROSPECT
The project is EU financed under the European Community’s Eighth Frame- work Program (Horizon2020). Its consortium is composed by seventeen part- ners made up of car manufacturers, research institutes and universities [2].
Collaborating partners are the following:
• Applus IDIADA
• Audi
• BMW
• Budapest University of Technology and Economics
• BAST
• Chalmers University of Technology
• Continental
• DAIMLER
• IFSTTAR
• TNO
• BOSCH
• VTI
• The University of Nottingham
• Toyota
• University of Amsterdam
• Volvo Cars
• 4a Engineering
1.1. BACKGROUND 3
Figure 1.1: The PROSPECT Logo.
The project started in May 2015 and the partners will work together during 42 months, during this time they will work for five common objectives.
Frequent meetings will be held during this time to ensure that the same method is used to reach the following objectives:
1. Studying different European accident databases and combining that with naturalistic observations will bring a better understanding to the PROSPECT project of accident scenarios with VRU.
2. To get better detection of VRU in different situations with larger sensor coverage, faster detection and better path recognition.
3. To implement more sophisticated Human Machine Interface HMI, and enlarge the parameters the vehicle will use in a situation for more pre- cise collision avoidance. For example, include steering in the AEB system.
4. For validation of the results from objectives 1 to 3 vehicle demonstrators are needed, they will be provided by Continental, Daimler, BOSCH and Volvo. There will also be a mobile driving simulator and a new realistic dummy for testing.
5. The final validation of the project will be realized with testing in realis- tic traffic scenarios including a study of the user acceptance. The final objective is to develop new test directives wit Euro NCAP.
4 CHAPTER 1. INTRODUCTION
1.1.2 Applus IDIADA
The Institute for Applied Automotive Research, IDIADA, started in the Uni- versity of Catalonia in 1971 as a small automotive research department. In 1990 IDIADA went to be an independent company 100% owned by the re- gional government of Catalonia. Applus bought 80% of the company in 1999 and now it is under the name Applus IDIADA.
Applus IDIADA has in a short time become one of the most important ser- vice, testing and engineer companies for the automotive industry. They work in 23 countries and have 47 local offices world-wide, the headquarter are in Santa Oliva, Tarragona, Spain. [3]. The services IDIADA offers are:
• Engineering
• Homologation
• Proving ground testing
• Test facility design
Figure 1.2: AEB system under test at Applus IDIADA1.
The department at Applus IDIADA that will work with the PROSPECT project is ADAS, ADAS is an abbreviation of Advanced Driver Assistance Systems and it is part of the Electronic Chassis Control Systems department.
The ADAS team is responsible for the development of systems like Active Cruise Control (ACC), Autonomous Emergency Braking (AEB) and Lane Change Assistance (LCA).
1Applus IDIADA, http://www.prospect-project.eu/wp-content/uploads/2015/
05/proactive-safety-pedestrians-cyclists-02-1200x600.jpg, January 2016
1.2. FUTURE BENEFITS 5
1.2 Future benefits
The future benefits of this project are many. The development of new active safety systems like more sophisticated AEB and AES systems. There are already autonomous driving vehicles, like the Tesla Model S, and the future leans at that direction. When we add this new parameters to the AEB system and making it more precise and reliable, we are coming closer to a future where autonomous driving vehicles are more common.
IDIADA’s and PROSPECT’s close collaboration with Euro-NCAP creates the opportunity to together develop new and more sophisticated crash tests.
To improve the crash tests it is important to develop new more realistic VRU dummies, this is only possible with a large amount of naturalistic driving data, NDD. At the moment, there are not an international standard for NDS of driver and VRU interactions. Due to the magnitude of the PROSPECT project, there is a possibility of creating a standard. A new standard and the creation of a database of the statistics produced from the results will simplify future investigations.
To sum it up, this thesis, as a part of the bigger PROSPECT project, will contribute to a future with more sophisticated AEB systems that hopefully lead to a future with decreasing VRU fatalities.
1.3 Project boundaries
This thesis covers just a part of the PROSPECT project, this project started in May 2015 and will last for 42 months. Due to the size of PROSPECT it is easy to extend the thesis. This gives clear boundaries a special importance.
One of IDIADA’s task is to do naturalistic observations and analyses of the collected data. They will start in January 2016 and finish in July 2016.
This thesis will be finished and presented by May 2016 and therefore it is impossible to cover all the collected data in this thesis. The first and most important boundary is that only the data collected to the middle of April will be analysed in this thesis. Driving the test vehicle in Barcelona requires special competence and it is time consuming, there will be 1000 hours of continuous recording. Therefore there will be professional drivers covering that. The installation of equipment will be supervised by employees from IDIADA An extension of this thesis will be the installation, calibration and validation of the equipment. There is a proposal of doing two reports due to the confidential material in the project. One report will be public and the other one will be confidential and only available to authorized people. There will also be a double presentation, one at IDIADA and one at the university, it
6 CHAPTER 1. INTRODUCTION is necessary due to the distance between the university and IDIADA. Both the two reports and the double presentation should be considered as an extension of the thesis.
Chapter 2 Theory
To get a good understanding of the different methods and definitions that are used in this kind of analyses, a profound theory study was accomplished.
Hereinafter the result of this study is presented.
2.1 Definitions
Theoretical definitions used throughout the thesis are defined here.
2.1.1 Time to Collision
TTC is the time left for the vehicle and the VRU to reach the theoretical collision point if both continue with the same velocity and direction. In general, a TTC value smaller than 1.5 s is defined as critical in urban areas and can lead to an accident. This value is called T T Cmin. D is the distance between the two objects, V 1 and V 2 the absolute velocity for each object.
T T C = D
V1− V2 (2.1)
7
8 CHAPTER 2. THEORY
Figure 2.1: Illustration of TTC between two objects.
2.1.2 Post Encroachment Time
PET is the measurement in time of near misses. It is the time between the moment the first object leaves the conflict zone and the moment the second object reaches the conflict zone. T 1, the time when the first object leaves the conflict zone. T 2, the time the second object enters the same conflict zone.
A = T 2 − T 1 (2.2)
Figure 2.2: Illustration of PET between two objects.
2.1. DEFINITIONS 9
2.1.3 Forward Collision Warning
FCW activates when the vehicle detects an obstacle that could cause an accident and warns the driver visually through a warning light and with sound [4].
2.1.4 Human Machine Interface
HMI is the interface between the machine and a person, for example, the FCW that warns the driver. In that case the HMI is the FCW and the audio-visual warning it produces for the driver.
2.1.5 Vulnerable Road Users
VRU are the group of road users that are the most vulnerable, these are pedestrians, cyclists and moped riders.
2.1.6 Absolute velocity
The absolute velocity is the actual velocity the object have, for example, the velocity of a vehicle driving on the highway.
2.1.7 Relative velocity
The relative velocity is the velocity difference between two objects, for ex- ample, when a vehicle overtakes another vehicle, the difference of velocity between the two vehicles are the relative velocity.
2.1.8 Yaw, pitch and roll
Yaw, pitch and roll are the three rotating movements the vehicle does around Z, Y and X, where the rotation around Z-axis is the yaw. The rotation around Y-axis is the pitch and the rotation around X-axis is the roll.
10 CHAPTER 2. THEORY
Figure 2.3: Yaw, pitch and roll illustration1.
2.2 Autonomous Emergency Braking
A big part of the accidents are caused by lack of time to react due to low visibility or suddenly appearing objects, for example, VRU. These accidents are normally at low speeds and do not usually cause serious injuries to the occupants of the vehicles, but they can cause serious injuries to VRU. The AEB system was introduced to avoid this kind of accidents.
Currently there are many different models of AEB systems in the market but the function is similar. They combine the information from the vehicle, such as absolute velocity and trajectory, with some type of system that de- tects obstacles in front of the vehicle. The most common systems are Radar (Radio Detection And Ranging), camera and/or Lidar (Light Detection And Ranging), the type of systems depends of the manufacturer. When the AEB systems detect a situation that could cause an accident and there is no ac- tion from the driver, it will, depending on if the AEB system have FCW or not, warn the driver. If there is not any reaction or the situation gets worse, the AEB system applies the brakes. The pressure applied to the brakes also depends on the severity of the situation and/or the manufacturer [5].
1Walter Stockwell, http://www.embedded.com/print/4379454/, visited on March 2016
2.3. TEST EQUIPMENT 11
Figure 2.4: AEB illustration2.
2.3 Test equipment
Due to internal reasons within the project, the delivery of the vehicle were delayed several weeks, a decision was taken to use a vehicle from IDIADA.
This decision was necessary for speeding up the process. The following part will cover the equipment installed in IDIADA’s vehicle, a Peugeot 3008 eHDI.
Vehicle
Model: Peugeot 3008 Year: 2013
Engine: 84,0kW, Diesel Kerb weight: 1626kg AEB system: None Lidar[6]
IBEO Lux 4(model 2010) Laser class: Class 1
2Euro NCAP, http://www.euroncap.com/en/vehicle-safety/
the-rewards-explained/autonomous-emergency-braking/, visited on January 2016
12 CHAPTER 2. THEORY Wave length: 905nm
Range: 200m (average distance) Numbers layers: 4 parallel layers Industrial Ethernet switch
This device can fusion up to 6 Ibeo LUX sensors, it is provided with 8 ports.
Synchronization unit
Synchronize the signals from the sensors.
IBEO ECU
Data fusion and object detection, can compare information with the vehicle CAN. Preliminary not connected.
PC[7]
Neousys Rugged Intel Ivy Bridge Compact In-Vehicle Computer.
Model: Nuvo-3100VTC.
Processor: Intel Core i7-3610QE (Ivy Bridge) 2.3 GHz.
Operating System: Microsoft Windows 7 Professional, 64-bit.
Memory: Wide-Temp DDR3 1600 SO-DIMM Memory - 8 GB.
Primary Storage: Transcend 370 mSATA SSD - 256 GB - [0H].
Additional Storage: 2.5” hot swap SSD hard disk.
Ethernet ports: 4 ports.
USB ports: four 3.0 ports and two 2.0 ports.
Camera[8]
Logitech Webcam C930e
Resolution: 1280 x 720 pixels, 1920 x 1080 pixels Frames per second: 30fps
Field of view: 90o Connection: USB 2.0 CAN
The speed and steering wheel angle will be extracted from the vehicles CAN.
GPS
The GPS are VBOX from Racelogic, speed and position is extracted from the VBOX.
Trigger
Button panel sending an analogue signal (5V) and marks the time when it
2.4. FIELD OPERATIONAL TEST 13 was actuated.
Vector CAN
Vector CANcaseXL is used to connect the CAN’s to the PC.
Figure 2.5: Layout of the equipment in the vehicle
2.4 Field Operational Test
FOT is a method for testing and analysing driver support systems The pur- pose is to see if the systems function as desired. This thesis will cover NDS
14 CHAPTER 2. THEORY (Naturalistic Driving Study), but the FOT method explained in the FESTA handbook is worth mentioning because it includes good recommendations for the installation of the test equipment and the procedure of the study. Rec- ommendations that are worth having in mind are for example, that avoiding wireless connection and only having physical connectors you can reduce the hardware problems with 80%. The equipment is a heavy user of energy and to be sure that no data will be lost, an emergency battery should be installed. Another important prevention of data loss is to have up to 50%
more space on the hard drive and making continued backups of the informa- tion [9]. The structure of the study that the FESTA handbook recommends are represented in the following figure.
Figure 2.6: Recommended NDS structure3.
2.5 DOCTOR-method
The first traffic observations were made in the 50’s and the first systemat- ically observations were made in 1966. During the 70’s and the 80’s the
3FESTA handbook, http://wiki.fot-net.eu/index.php/FESTA_handbook_
Introduction, last updated 1 of April 2014, visited on February 2016
2.6. ANALYSIS METHOD 15 evolution of different observation methods really toke off. The Dutch Objec- tive Conflict Technique for Operation and Research, DOCTOR, standardised different observation techniques and made a well-defined manual, the DOC- TOR manual. Published the first time in 1986, DOCTOR defines different concepts, such as encounter, critical situation, conflict, evasive action...etc.
Hereinafter, definitions of these concepts will be explained.
• Encounter, is when two or more road users gets closer and their ac- tions can influence in each other.
• Collision course, is when one or several road users have to change their course to avoid a collision.
• Critical situation, appears when the time to react gets smaller, as defined before, when the TTC gets under 1.5 s the situation is consider dangerous in a urban area.
• Conflict, is a situation when two or more road users are approaching each other. And if they do not change their course or velocity there will be a risk of a collision and probably personal and material damage.
• Evasive maneuver, appears when the road user accelerate, brake or steer too much in a critical situation. This maneuvers can make the situation worse.
The DOCTOR manual categorize many different encounters in different types of conflicts, such as Car to Car, Pedestrian to Cyclist, Car to Pedestrian, Car to Cyclist...etc. The steps to identify a conflict are, detection of the conflict, determine the severity, probability of collision and extent of the consequences. The severity of the situation is divided in a scale of 1 to 5, to decide the severity of the situation it is necessary to have in mind the absolute and relative velocity and the type of road users. The probability of a collision is depending on the TTC or PET value. The extent of the consequences are depending on the kinematics and the reaction of the driver and VRU [10]. If there are more data available, the estimation of the conflicts severity will be more precise.
2.6 Analysis method
The method used to analyse our NDD (Naturalistic Driving Data) is a variant of the DOCTOR-method. The main difference is the time and the precision of
16 CHAPTER 2. THEORY the data. But also the procedure changes a bit. According to the DOCTOR- method, the time necessary for the observations are three days, we will do our NDS during 4 months. The reason to this difference is that in the DOCTOR- method there is a person standing in one intersection as an observant. Our NDS will be done from a vehicle that is driving around in the city and there will be much more variation in the situations, therefore it is necessary to record for a much longer time. Due to the equipment in the car, I will be able to extract more data from our NDS than what the DOCTOR-method covers [10]. For example, instead of guessing the velocity and position, we can extract that directly from the recorded data.
We will define two types of situations, conflict and close encounter. Close encounter is when no avoidance maneuver is needed to avoid an accident, but the situation is risky. The severity of the situation will be defined with the following parameters:
• Proximity
• Trajectory
• Speed
• Head orientation
• Risky behaviour
• TTC or PET
• VRU vulnerability
Chapter 3
Preparation and Execution of Analysis
A research of the vehicle equipment was necessary before starting the prepa- ration of the vehicle and analysis platform. It was also important to get a good understanding of the PROSPECT project and the collaborating part- ner’s tasks. There is several analysis methods that had to be studied and the most relevant are explained in Chapter 2.
3.1 Vehicle preparation
The delivery of the vehicle and the test equipment was delayed due to internal reasons within the project. Therefore, the preparation of the vehicle had to start much later than planned. To speed up the process, it was necessary to use a vehicle and equipment from IDIADA.
17
18 CHAPTER 3. PREPARATION AND EXECUTION OF ANALYSIS
Figure 3.1: The vehicle used in Barcelona, a Peugeot 3008.
3.1.1 Mounting of the test equipment
The vehicle from IDIADA already had some of the equipment mounted from older projects, so the mounting modifications were already made. First the Lidar was mounted, it is placed in front of the vehicle, in the grill. It is sup- ported by a metal bracket that gives adjusting possibilities for compensating of the vehicles pitch angle. A Plexiglas was placed in front of the Lidar to protect it from dirt.
Figure 3.2: The Lidar.
The Lidar will give us information about the position of the VRU and
3.1. VEHICLE PREPARATION 19 the relative velocity between the vehicle and the VRU. It is connected to the synchronization unit via a synchronisation cable, and to the industrial Ethernet switch with a LUX Ethernet cable.
Figure 3.3: Ethernet switch and synchronization unit.
The industrial Ethernet switch is connected to the PC, via Ethernet. To the PC there are also connected two cameras (with USB), an external display (through VGA) and a Vector CAN, via USB. The cameras are attached to the windscreen with a sucker. To ensure that it would not fall down, tape was added.
Figure 3.4: The camera pointing to the road.
20 CHAPTER 3. PREPARATION AND EXECUTION OF ANALYSIS The Vector CAN is connected to the vehicle’s CAN, via a serial port and extracts the velocity and steering wheel angle.
Figure 3.5: The VectorCAN to the left, USB-hub and the trigger to the right.
The VBOX is connected to the Vector CAN using a serial port. It has an external GPS receiver placed outside the vehicle.
Figure 3.6: VBOX.
3.1. VEHICLE PREPARATION 21 The trigger is also connected to the Vector CAN via a serial port, the trigger is actuated via a keypad placed where the gear leaver.
Figure 3.7: Keypad for the trigger.
The equipment has a 12VDC supply from the supply box. The supply box is connected to the car’s battery via a fuse box with a switch.
Figure 3.8: Fuse box with a switch, hard drive mounted on top.
22 CHAPTER 3. PREPARATION AND EXECUTION OF ANALYSIS The PC, Vector CAN, Ethernet box, synchronization box, battery switch and feeding box are placed in the trunk. The equipment that is placed in the trunk have ventilation from four fans, connected to the 12VDC feeding box.
All the equipment is fixed safely with velcro. The cables are laid as discrete as possible and fixed with tape and cable ties.
Figure 3.9: Cooling system.
3.1.2 Lidar configuration
When all the equipment was installed and before the data acquisition begun, the Lidar had to be calibrated and configured. To start with, the vehicles dimension had to be inserted, such as length, distance between front to front axle and distance between axles. The position of the Lidar had also to be introduced, in this case in the front middle of the vehicle. The calibration of the Lidar was effectuated with a rectangular box with the same height as the bottom of the Lidar. The box is placed at 10m of distance just in front of the vehicle on a levelled surface. IBEO’s software were used to visualize the four layers of laser from the Lidar. Each layer has its own colour, so it is easy to difference the two top layers from the two down layers. One driver and one passenger had to be in the vehicle to have the correct pitch and roll angle. The Lidar was adjust until we could see that the two down layers was reflected from the box, and the two upper layers did not.
3.1. VEHICLE PREPARATION 23
Figure 3.10: Lidar configuration setup1.
1. Vehicle 2. Lidar
3. Levelled surface 4. Optical axis
5. Laser detector (Not used) 6. Reference target
3.1.3 Validation
The last part of the preparation of the vehicle was to test and verify that the equipment works. I was present during the two first days of recording in Barcelona to check that everything was working as it should. During the first days it appeared problems with the Start/Stop system from the car that created problems for the equipment.
1Operating Manual ibeo LUX 2010 Laserscanner, version 1.6R
24 CHAPTER 3. PREPARATION AND EXECUTION OF ANALYSIS
Figure 3.11: Layout for controlling.
During the first weeks of driving we had to use a temporary PC due to a delay of the In-vehicle PC. The temporary PC caused some connectivity problems due to its lack of performance. The connection problem with the Lidar caused an error message every hour, and a restart of the recording was necessary every the error message appeared. Due to the lack of space in the hard drive an external hard drive was necessary. At first it was connected via eSATA and it lost connection several times during the recording. Changing from eSATA to USB solved the connectivity problem with the hard drive.
3.2 Designing of the analysis platform
The project had already commenced when I incorporated and started the thesis. So the main part of the parameters were already proposed and sum- marized in a Excel document. But these parameters were only preliminary and had to be discussed before designing the analysis platform. Since the analysis is done in parallel with IFSTTAR in Lyon and Budapest University of Technology and Economics. It is especially important that the results presented have the same parameters, structure and design. Therefore, a two days meeting was held at IFSTTAR in Lyon the 25 and 26 of January. Dur- ing these days the definitive parameters that should be used in the analysis were decided and an analysis platform was created in Excel.
There will be a big amount of data, so there is a need to create a data-base.
This is possible to do in several programs, such as Access, Matlab and SQL.
This database will only be used at IDIADA and does not have to be similar to the ones at IFSTTAR and BME. However the presentation of the results
3.3. ANALYSIS PROCEDURE 25 and statistics will be combined with the results from IFSTTAR and BME, so they need to have the same layout.
3.3 Analysis procedure
The delayed vehicle delivery also affected the start of the NDS, which in turn affected the start of the analysis of the NDD.
The amount of data produced is big, one day of recording is around 100GB.
All this data is extracted from the computer in the vehicle once a day and stored in a NAS (Network-attached storage). The files are organized by date, morning or evening shift and hour. For every day there is an Excel document where additional information can be added, for example, if there have been an incident or problems with the equipment. This makes the analysis more efficient.
The data was analyzed during one month using the vADASdeveloper from Vector. It is possible to visualize the two cameras and the Lidar at the same time, as shown in the picture below.
Figure 3.12: The analysis software vADASdeveloper.
Vehicle CAN data and VBOX data can also be visualized on the same application.
As the Lidar give us XY positions of the objects with respect to the rear axle of the vehicle it is possible to calculate the relative velocity and the relative position. X is perpendicular to the rear axle increasing longitudinal to the
26 CHAPTER 3. PREPARATION AND EXECUTION OF ANALYSIS front and Y is parallel to the rear axle increasing to the left. A negative relative speed means that the distance between the vehicle and the VRU is increasing, and no critical situation will occur.
RelativeSpeed =
q
x21+ y12−qx22+ y22
t12 (3.1)
When the relative velocity and position are known the TTC is easy to calculate. The absolute velocity, acceleration and the trajectory of the VRU is not possible to calculate due to the constantly moving coordinate system that has the origin on the vehicles rear axle.
Chapter 4 Results
After the vehicle was equipped and the data from Barcelona collected and analysed it was time to get some results from the work done. Hereinafter you will find the resulting layout of the vehicles equipment and the statistics from the analysis.
4.1 Vehicle equipment
When the new In-vehicle PC was installed and all the equipment started to work correctly, the equipment was secured and the wiring was cleaned up.
Figure 4.1: Final equipment layout with the new In-vehicle PC.
27
28 CHAPTER 4. RESULTS The VBOX that was installed provided the velocity and coordinates of the vehicles position. However due to the infrastructure in Barcelona which causes problems with the connection to satellites, it resulted not to be reliable to only use the GPS for vehicle speed. As the vehicle’s CAN was used to extract the vehicle data, it was possible to extract the velocity from the CAN.
4.2 Driver interface in the vehicle
A driver instruction was made for the drivers to have possibility to solve some of the software problems that usually appear. It also includes instructions for the start-up of the equipment and the most important configurations specifications for the equipment, cf. appendix C. The camera were mounted so that they would interfere as little as possible to the driver’s road visibility.
The screen was mounted in front of the passenger seat and in a position that would be easy for the driver to check the status when the vehicle is standing still.
Figure 4.2: Overview of the drivers interface.
4.3. ANALYSIS PLATFORM 29 Due to the problems that the Start Stop system of the vehicle gave, it was necessary to put a reminder stick on the dashboard to turn the ECO-mode off.
Figure 4.3: Eco off sticker.
4.3 Analysis platform
The final collection of parameters was defined during the meeting in Lyon.
These parameters are summarized in an Excel document that later on will be used as the analysis platform, this can be seen in Appendix B. The pa- rameters are divided in four groups.
1. The first one is called Global and covers weather, time, infrastructure and VRU type.
2. The second defines the encounter of the VRU and the vehicle, for ex- ample, type of encounter and yielding.
3. The third explains the head/torso orientation, and gestures, this part is called Intents.
4. The last part is the Kinematics and explains the velocity, acceleration and positions.
30 CHAPTER 4. RESULTS The GIDAS (German In-Depth Accident Study) pictogram’s are used to determine the type of encounter, the different encounters are enumerated from 0 to 17. They can be seen in Appendix B.
The head and torso orientations were decided to have different precisions, since it is more difficult to determine a precise torso orientation than the head orientation. The head orientation is with a precision of 16 different positions and the torso orientation with 8 positions.
At this stage the data-base that was used is just a copy of the analysis platform.
4.4 Analysis results and statistics
The analysis resulted in 95 hours of video analysed and 1.46TB of data.
This produced 84 critical situations that were analysed and coded in an Excel.
These critical situations are the base for the statistics presented hereinafter in form of graphs and tables. Only the most interesting from a sensor developing point of view are presented.
4.4.1 Global statistics
The first graph is the VRU type distributions of which type of VRU that creates most critical situations. Alternative mode for pedestrian refers to a pedestrian on a skateboard, roller blades or something similar. The Pedes- trian carrying something is a pedestrian with a dog or with a trolley.
4.4. ANALYSIS RESULTS AND STATISTICS 31
Figure 4.4: Pie chart of the type of VRU.
As can be seen in the graph and the table, the pedestrians are the type of VRU that most commonly appear in a critical situation.
Table 4.1: VRU type distribution
VRU type Total
Alternative mode for pedestrian 1
Cyclist 11
Pedestrian 67
Pedestrian carrying something 5
32 CHAPTER 4. RESULTS
4.4.2 Encounter statistics
The right of way is an interesting parameter. This is where the conflict usually starts, when someone breaks a traffic rule. Absolute refers to right of way without condition. Conditional, the right of way depends on the right of way of other road users. Not permitted, has not right of way. As seen in the plot below, the big part of the conflicts are caused by a VRU braking a traffic rule.
Figure 4.5: Column plot of the right of way.
In the table below it can be seen thee exact number of the right of way of driver and VRU.
Table 4.2: Right of way Right of way Driver VRU
Absolute 72 4
Conditional 11 14
Not permitted 1 66
4.4. ANALYSIS RESULTS AND STATISTICS 33 The graph below presents the type of encounter recording according to the GIDAS pictograms.
Figure 4.6: Column plot of the type of encounter recording to the GIDAS pictogram’s.
Two types of encounter stand out from the others, number 4 and 8. They are explained in the table below. More detailed description with illustrations of the GIDAS pictogram’s can be found in Appendix B.
34 CHAPTER 4. RESULTS Table 4.3: Type of encounter
Number
Scenario Scenario description of
situations
0 Not defined in GIDAS 1
4 Straight on from right, no intersection 35
6 Straight on from left on intersection 3
7 Straight on from left from cycle lane 1
8 Straight on from left, no intersection 25
10 Turns in to oncoming traffic, not crossing 4
11 Turns in to oncoming traffic, crossing 3
12 Turns in to oncoming traffic, crossing after intersection 4
14 Longitudinal 1
15 Longitudinal 3
16 Longitudinal 4
As seen above the most common encounter is when the pedestrian is crossing straight from the right or straight from the left where there is no intersection. This results is what were found in the literature study of ac- cident databases. It is also worth mention the longitudinal cases that have important information for development of AES systems.
4.4. ANALYSIS RESULTS AND STATISTICS 35
4.4.3 Intent statistics
The head orientation is an important parameter from a sensor development point of view. As shown in the radar graph below, the head orientation angle of 90o is the most common.
Figure 4.7: Radar plot of the VRU head orientation respect the vehicle.
As seen in the plot, in a big part of the cases the head is oriented to the vehicle (270o to 90o) This usually means that the pedestrian sees the vehicle.
To better understand the plot above, the following figure is the one used as a reference during the analysis of the head orientations.
36 CHAPTER 4. RESULTS
Figure 4.8: Figure used during the analysis of the head orientation.
Table 4.4: Head orientation distribution Head orientation VRU Number of situations
0o 8
22.5o 7
45o 6
67.5o 6
90o 20
112,5o 2
135o 3
157,5o 1
180o 2
202,5o 2
225o 3
247,5o 1
270o 7
292,5o 3
315o 8
337,5o 4
4.4. ANALYSIS RESULTS AND STATISTICS 37 The torso orientation is also an important parameter from the sensor development point of view. As shown in the radar graph below, the torso orientation angle of 270o is the most common. 270o in torso orientation is equivalent to 90o in head orientation. This way of defining the orientations is due to the need of it from a collaborating partner, and the need of homo- geneity of the parameters between the partners.
Figure 4.9: Radar plot of the VRU torso orientation respect the vehicle.
38 CHAPTER 4. RESULTS To better understand the plot above, the following figure is the one used as a reference during the analysis of the torso orientation.
Figure 4.10: Figure used during the analysis of the torso orientation.
Table 4.5: Torso orientation distribution Head orientation VRU Number of situations
0o 6
45o 10
90o 13
135o 7
180o 1
225o 8
270o 31
315o 7
As seen both in the hed and the torso orientation, the majority of the cases is around 90o and 270o. This make senses when comparing to the GIDAS pictograms where the majority of the cases is crossing perpendicular to the vehicle.
4.4. ANALYSIS RESULTS AND STATISTICS 39
4.4.4 Kinematic statistics
The speed combined with TTC and PET are the most important parameters for defining the severity of the conflict. The speed was read at the lowest TTC in the case of having TTC. For the cases without TTC the speed was read just before the driver start the maneuver. The box plot below shows the absolute vehicle speed.
Figure 4.11: Box plot of the vehicle speed.
The table below shows the different statistical variables of the box plot.
Table 4.6: Statistical values of the vehicle speed Statistical variable Value
Max 38.00 km/h
Mean 18.77 km/h
Median 16.50 km/h
Min 4.00 km/h
Standard deviation 7.83 km/h
40 CHAPTER 4. RESULTS The pie chart below shows the distribution of TTC and PET in the dif- ferent situations. As shown in the graph, the PET is the most common, but there are also nine cases where neither TTC nor PET can be calculated.
These cases are normally when the VRU comes in an angle out of the field of view for the Lidar.
Figure 4.12: Distribution of the TTC and PET.
23 cases reported a TTC which means that an action had to be taken by the driver to avoid a more critical situation. This is when an AEB system should brake. 52 cases reported a PET which means that no action by the driver is needed. This is the most difficult part for an AEB sensor point of view. This is when the vehicle should not brake.
4.4. ANALYSIS RESULTS AND STATISTICS 41 The TTC plot shows the most common TTC per situation for the 23 situations that reported TTC. In the plot there is a reference line indicating the TTCmin of 1.5 s. The majority of the situations with a TTC are to the left of this line and therefore have a TTC under 1.5s, only two cases are to the right of the line.
Figure 4.13: Plot of TTC.
The exact number of situations per TTC can be seen in the table below.
1.4 s stands out of the crowd as the most common TTC with 8 situations with this TTC.
42 CHAPTER 4. RESULTS Table 4.7: TTC distribution
TTC (s) Number of situations
0,5 1
0,7 1
0,8 3
1 1
1,1 3
1,2 2
1,3 2
1,4 8
1,7 1
2,8 1
The PET plot below shows the most common PET per situation for the 52 situations that reported PET. The majority of the situations with PET are 1s or lower. PET does not have a definition of the PETmin, but a PET under 1.5 s can also be seen as a critical situation.
Figure 4.14: Plot of PET.
4.4. ANALYSIS RESULTS AND STATISTICS 43 The exact values of the numbers of situations per PET can be seen in the table below. 1 s stands out of the crowd as the most common PET with 15 situation with this PET. It is a quite low value and could in a AEB sensor point of view be difficult to calculate if it is necessary for the vehicle to brake or not.
Table 4.8: PET distribution PET (s) Number of situations
0,5 2
0,6 1
0,7 1
0,8 5
0,9 7
1 15
1,1 4
1,2 2
1,3 3
1,4 4
1,5 2
1,8 1
1,9 1
2 3
2,2 1
44 CHAPTER 4. RESULTS
Chapter 5 Discussion
The results of this thesis should be considered reliable. But as with most things, everything can be done in a different way and be improved.
The biggest setback has been the non-delivery of the vehicle with equipment promised to be at IDIADA the last week of January. The delivery date have been moved forward several times and the 7 of March 2016, 5 weeks later, we decided to use a vehicle from our department at IDIADA. This gave me the opportunity to include more steps of the Naturalistic Driving Study in the thesis, such as the mounting, configuration and calibration of the equipment in the vehicle. But, I could only analyse during four weeks, not during the ten weeks that was the preliminary objective. Ten weeks instead of four would have given me more reliable statistics and more cases with high severity. The vehicle would have been equipped with a more sophisticated equipment, for examples, an automatic trigger. With a manual trigger as in our vehicle, could give a human error to the data collected. With an automatic trigger choosing the critical situations we could eliminate that error. There have also been other setbacks, a punctured tire stopped the data acquisition in Barcelona for five days. After all, including the setbacks, as said before, the results presented in this thesis should be considered reliable.
As the vehicle will be collecting more data during some more weeks, there will be a need to continue analysing data. I got the trust from IDIADA to continue this work using the methodology described and performed in this thesis.
And not to forget, being at Applus IDIADA during this 19 weeks have given me a perspective of the automotive engineering that I doubt I could get from another company.
45
46 CHAPTER 5. DISCUSSION
Chapter 6 Conclusion
The conclusion from the statistical results presented in this thesis is clear.
There is one type of situation that stands out in the analysis. The most common case is when a pedestrian breaks a traffic rule, crosses the road perpendicularly and creates a PET (Post Encroachment Time) of 1 second.
This is an ideal type of conflict that can be solved with an AEB (Autonomous Emergency Breaking) system and in the future combining it with AES (Au- tomatic Emergency Steering).
The vehicles mean speed is 18.77 km/h. Today’s AEB systems can avoid complicated situations up to 20 km/h and avoid situation without obstruc- tion with speeds over 60km/h [11]. This is already good results, but we had cases with speeds up to 38 km/h with obstruction and in some situations the driver also needed to turn to avoid an accident. Therefore we need data as the one presented in this thesis for future developing.
47
48 CHAPTER 6. CONCLUSION
Chapter 7
Further work
An improvement would be to speed up the process of analysis. Create a software that saves the standard parameters, such as the speed and time.
This process should be done automatically, saving the data in to a database during the driving. This would improve the efficiency of the analysis and make it easier to analyze large amounts of data.
The scope of this thesis is to acquired and analyze data. A recommendation would be to go further and use this data to develop conditions and parameters for Automatic Emergency Steering (AES). This systems does already exist, but they are not standardized. In 2020, Euro-NCAP want to include AES systems and it is important to have a standard by then [12].
49
50 CHAPTER 7. FURTHER WORK
Appendix A Project Plan
51
Project Plan
Milestones
Report writing
The report is a document that will be built continuously from the start in January till the end in May. To ensure quality and enable active support of my examiner Jan van Deventer, the report will be written in LaTeX. With support of my supervisor Laura Sanz and the project manager Andrés Aparicio I will ensure that no confidential material reaches the report. There is also a propose of making two reports, one confidential and one public.
Pre studies
The first preparations of this thesis are to do an opposition of a previous thesis and develop this project plan. To get a better understanding of the PROSPECT project a profound literature studies will be performed.
Design of data acquisition software
Available software for data analysis is used, but we have to design our own method and develop the parameters we want to use. Before starting, there will be a meeting at IFSTTAR in Lyon to decide a common method for data analysis. IFSTTAR and IDIADA will be analysing data parallel, so it is very important to use the same method to get reliable results.
Equipment preparations
The preparation of the equipment and the vehicle is carried out by employees at IDIADA, my part will be observing to get a clear view of how the equipment will work in the vehicle. This includes also a profound information search of the different ingoing components.
Data acquisition
The data acquisition is not my responsibility, professional drivers will driving the vehicle in Barcelona during 4 months starting in February. But it is worth mentioning because the whole thesis depends on the data acquired I Barcelona. I will not use all the data to this thesis due to the lack of time.
Analysis of naturalistic driving observations
It is necessary to fusion the different parameters, such as speed, GPS position, drivers camera, road camera…etc. A Global analysis will be applied to all situations, for example lighting conditions, precipitation and traffic density. The critical situations will have a more profound analyse, for example head orientation, the VRU’s gestures and torso orientation. The
information from the analyses will be used to produce statistics. The final stage of the analysis is to create a conclusion of the statistics.
Presentation
The development of the presentation will start at the final part of the thesis. Due to the distance between the company and the university, two presentations is needed, one for IDIADA and the final presentation for the university.
Project boundaries
My thesis covers just a part of the PROSPECT project, this project started in May 2015 and will last for 42 months. Due to the size of PROSPECT it is easy to extend the thesis. This gives clear boundaries a special importance. One of IDIADA's task is to do naturalistic observations and analyses of the collected data. They will start in January 2016 and finish in July 2016. This thesis will be finished and presented by May 2016 and therefore it is impossible to cover all the collected data in this thesis. The first and most important boundary are that only the data collected to the middle of April will be analysed in this thesis. Driving the test vehicle in Barcelona requires special competence and it is time consuming, there will be 1000 hours of continuous recording. Therefore there will be professional drivers covering that. The
installation of equipment will also be carried out of employees from IDIADA, but an extension of this thesis will be covering the ingoing parts of the equipment and how they are installed in the vehicle. There is a proposal of doing two reports due to the confidential material in the project. One report will be public and the other one will be confidential and only available to authorized people. There will also be a double presentation, one at IDIADA and one in the university, is necessary due to the distance between the university and IDIADA. Both the two reports and the double presentation should be considered as an extension of the thesis.
Gantt chart
Start End 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Report writing 2 20
Pre studies 1 4
Opposition of thesis Falkenberg / IDIADA 1 2
Project plan IDIADA 2 3
Literature studies PROSPECT IDIADA 2 4
Design of data acquisition software 4 7
Meeting with project partners IFSTTAR 4 4
Develop parameters IDIADA / IFSTTAR 4 7
Determine analysis method IDIADA / IFSTTAR 4 7
Equipment preparations 5 7
Information retrieval of car equipment IDIADA 5 6
Install equipment in car IDIADA 5 6
Data acquisition 7 24
Naturalistic driving observations Barcelona 7 24
Analysis of naturalistic driving observations 7 19
Fusion of the different parameters IDIADA 7 15
Analysis of critical situations IDIADA 7 16
Statistics of the different situations IDIADA 12 17
Conclusion IDIADA 17 19
Presentation 19 21
Prepare presentation IDIADA / LTU 19 21
Final presentation company IDIADA 21 21
Final presentation university LTU 21 21
Locations Abbreviations
Falkenberg, Sweden Falkenberg
Applus IDIADA, Santa Oliva, Spain IDIADA Luleå University of Technology, Luleå, Sweden LTU
IFSTTAR, Lyon, France IFSTTAR
Barcelona, Spain Barcelona
Week
Chart Location
Risk analysis
Daniel Castro Larsson Luleå University of Technology
Risk Probability (1-10) Impact (1-10) Risk score (PxI) Solution
Lack of literature 1 4 4
More information search and good communication with IDIADA and partners
Bad communication between partners 2 6 12
More meetings and communication course
Lack of information about the equipment 3 3 9
Information search and contact equipment manufacture
Delay of vehicle delivery 5 5 25Access to a spare
vehicle at IDIADA
Vehicle failure/accident 2 8 16Access to a spare
vehicle at IDIADA
Failure of equipment 3 8 24
Access to spare equipment at IDIADA
Lost of data 4 8 32
Backup of all information acquired and all the documents
Lack of motivation 1 7 7
Good planing and structure. Be motivated, it is fun
Lack of time 4 6 24Good planing and
structure.
Disease 2 6 12
Contact medic as son as possible,
alternative, work from home
56 APPENDIX A. PROJECT PLAN
Appendix B Parameters
A reduced version of the parameters divided in Global, Encounter, Intent and Kinematics. Including also graph of head/torso orientation and GIDAS pictogram’s. This document belongs to Applus IDIADA, IFSTTAR and Budapest University of Technology and Economics as part of the project PROSPECT (Horizon 2020 Grant Agreement No. 634149). A full version of the parameters will be presented in the deliverable of the PROSPECT project ”D2.1 Accident Analysis, Naturalistic Driving Studies and Project Implications”.
57
