Vous êtes sur la page 1sur 6

2015 IEEE Intelligent Vehicles Symposium (IV)

June 28 - July 1, 2015. COEX, Seoul, Korea

Impact of Positioning Uncertainty of Vulnerable Road Users on Risk


Minimization in Collision Avoidance Systems*
P. Themann, J. Kotte, D. Raudszus and L. Eckstein


Abstract This work describes a methodology to assess the


impact of positioning and prediction accuracy on the potential
benefit of collision avoidance systems. The predicted position of
vulnerable road users (VRU) ahead of the vehicle is affected by
measurement and prediction uncertainty. In advanced
cooperative collision avoidance systems the position of VRUs is
provided by vehicle-to-vehicle or vehicle-to-infrastructure
(V2X) communication. This work describes a method to

         l trajectories in
critical situations in order to minimize the risk of the situation
considering the influence of positioning and prediction
inaccuracies of VRU. The findings discussed here define
requirements on the prediction accuracy and for vehicle
velocities of 50 km/h the predicted VRU position should
provide a standard deviation of less than 55 cm.

I. INTRODUCTION AND MOTIVATION


Pedestrians account for a significant portion of road
traffic fatalities. According to WHO [1] 273,000 pedestrians
were killed in road traffic in 2010 worldwide, which
represents 22% of all road traffic deaths. Passive safety
measures reduce injury severity in case of a collision. In
contrast, active safety systems aim to avoid or mitigate
collisions by means of braking or steering interventions or by
providing warnings to the driver [2]. In many situations
pedestrians cannot be detected by vehicle sensors because of
obstructions or a limited field of view and hence cooperative
approaches show potential to widen the impact and increase
safety [3]. In order to support the development of such
communication-based safety systems, the project CERM
(Center for European Research on Mobility) has been
initiated at RWTH Aachen University. Within this project, a
test intersection is being build up that allows assessing V2Xsystems at interactions with surrounding traffic and road side
units as well as VRUs. One of the goals of CERM is to
improve VRU safety by utilizing Smartphone position data,
which are provided to vehicles by radio communication. As
the effectiveness of these systems significantly depends on
the accuracy of the predicted VRU position, which is affected
by positioning sensor accuracy and the quality of prediction,
this work presents an approach to assess the impact of an
uncertain VRU position on the potential of a communicationbased safety system avoiding collisions by evading and
braking.
*Research supported by the project UR:BAN and the CERM initiative
P. Themann manages the development of ADAS at the institute for
automotive engineering (ika), RWTH Aachen University, (phone: +49-24180-25673; e-mail: themann@ika.rwth-aachen.de).
D. Raudszus and J. Kotte are with the ADAS department at the institute
for automotive engineering (ika), RWTH Aachen University, (e-mail:
{raudszus, kotte}@ika.rwth-aachen.de).
L. Eckstein is the head of the institute for automotive engineering (ika),
RWTH Aachen University, (e-mail: eckstein@ika.rwth-aachen.de).
978-1-4673-7266-4/15/$31.00 2015 IEEE

II. COLLISION AVOIDANCE AND MITIGATION SYSTEMS


A. State of the art system approaches
Collision mitigation systems currently available on the
market detect surrounding objects based on environmental
sensors, perform a prediction of object position, assess
situation criticality and perform an appropriate action, as e.g.
described in [4]. Known approaches use e.g. time to collision
(TTC) [5] time to brake (TTB) or time to steer (TTS) [6] as
$3*5*$"-*5:.&"463&48)*$)*.1-*&45)"50#+&$5@4.05*0/45"5&4
are nearly constant. These approaches prove to successfully
address rear end vehicle collisions. However, such
deterministic criticality criteria are not appropriate for
pedestrian collision mitigation systems, since acceleration
and direction of pedestrian velocity are subject to highly
dynamic changes [7]. In order to cover non-deterministic
pedestrian movement, a probabilistic approach for criticality
assessment is proposed in [8]. This considers uncertainty of
both pedestrian position and predicted actions which allows
for integrating a probabilistic pedestrian behavior model as
described in [9]. In case of a critical situation, de-escalation
strategies are implemented that contain required braking [5]
or steering actions [10] to avoid or mitigate an imminent
collision.

Figure 1. Injury risk curves based on GIDAS accident data [12]

B. VRU Injury Risk


According to ISO 26262 [11], risk is defined as
=$0.#*/"5*0/ 0' 130#"#*-*5: 0' 0$$633&/$& 0' )"3. "/% 5)&
severity of that )"3.?910463&%&1&/%40/5)&'3&26&/$:0'
collision, which is available from accident data bases, and
effectiveness of the collision mitigation system that might
avoid a portion of collisions. Severity mainly depends on
collision speed, collision point and passive safety measures
of the colliding vehicle. Within the EC funded project
AsPeCSS injury risk curves have been developed that

1201

describe the impact of collision speed on the probability of


injured casualties in pedestrian accidents based on an analysis
of GIDAS (German in-depth accident study) accident data
[12]. The risk curves for fatally, seriously and slightly injured
pedestrians are depicted in Figure 1.
III. PREDICTION OF VRU BEHAVIOR
The prediction of VRU behavior is affected by several
inaccuracies that are discussed in the following sections.
A. Assessment of accuracy of positioning systems
Positioning systems are subdivided into local and global
systems. A local positioning system is e. g. the RadioFrequency Identification (RFID) based system proposed in
[13]. This kind of systems provides position accuracies better
than 10 cm [13]. A major disadvantage of this kind of
systems is that it requires an adaption of the environment.
Contrary most commonly used Global Navigation Satellite
Systems (GNSS) do not require an adaptation of the
infrastructure, but only provide limited accuracies. One
example for GNSS with a high utilization rate in Europe and
the USA is the Global Positioning System (GPS).
Measurements of GPS position accuracy e.g. in [13] reveal
that inaccuracies in position of 10 m have to be expected. The
most influential source for GNSS position inaccuracies are
ionosphere errors. To reduce global position errors and
improve the accuracy, augmentation systems like Integrated
Navigation Systems (INS), Differential Global Navigation
Satellite Systems (DGNSS) and Space Based Augmentation
Systems (SBAS) were developed. With augmented systems
like these, position accuracies up to 1-3 cm are possible
[14]. In [15] an additional augmentation system, which uses
3D city models to reduce shadowing effects of GPS signals,
is described. The proposed system can improve position
accuracy of a standard GPS system from 10 m to 5 m. The
European GNSS Galileo is designed to provide a positioning
accuracy of up to 4 m in a free of charge mode and an
accuracy of up to 1 m in a commercial mode [16].
B. Integration of sensors from consumer electronics devices
Consumer Electronics with integrated acceleration
sensors, gyroscopes, GNSS and communication services can
also build the basis for a positioning system. It is also
possible to receive DGPS information and thus to improve
the position accuracy to less than 1 m [17]. Next to
augmentation GNSS, it is also possible to estimate the
position by a cross bearing of reachable mobile radio stations
[18].
C. Integration of infrastructure-based sensors
Other approaches utilize infrastructure-based sensors to
detect VRUs. Sensor information is provided to vehicles by
means of road side units. Such systems have been
implemented in the research projects Intersafe2 [6] and
KoFAS [19] in the form of an intersection, which is equipped
with multiple laser scanners. Although infrastructure-based
sensors provide advantages compared to vehicle-based
4&/4034&(%&5&$5*0/0'0#+&$540654*%&5)&7&)*$-&@4'*&-%0'
view, the available information is still subject to uncertainty
due to sensor characteristics. However, suitable laser scanner
systems provide a position accuracy of below 10 cm, which
is significantly better than GNSS.

D. Prediction based on average behavior


Conventional Collision Avoidance Systems mainly
predict the VRU position by an extrapolation based on the
assumption of a constant acceleration of the VRU [9]. Since
amount and direction of pedestrian velocity are subject to
highly dynamic changes [7] these approaches lead to
unrealistic position estimations. To improve the prediction,
extended approaches build possible movement areas for
VRUs based on physiological movement possibilities. In
combination with movement patterns it is thus possible to
determine probabilities for predicted VRU positions. An
example for an approach like this is given in [20]. In this
approach the area, which can be reached with the current
velocity and yaw rate, is weighted with the highest rates.
Areas for different movement patterns of a VRU and a
horizon of 2 seconds are shown in Figure 2. Assuming the
VRU moves at a constant velocity (highest rates in [20])
without changing the direction the VRU travels 3.2 m. If the
VRU decides to accelerate to vmax or decelerate to vmin this
results in distances of 5.7 m or 1.2 m. Here a position
uncertainty of up to 4.5 m has to be considered. Changes in
the direction of movement further increase the uncertainty.

Figure 2. Areas for the movement patterns constant velocity, acceleration


to vmax (4.4 m/s) and deceleration to vmin (0 m/s) of a VRU with 1.6 m/s and
a horizon of 2 sec [20]

IV. RESEARCH METHODOLOGY


Prediction models for the dynamics of VRUs are affected
by inaccuracies and consequently the position of a VRU in
front of a vehicle cannot be assessed without the influence of
uncertainties. This work assesses the impact of positioning
inaccuracy on the potential benefits of two collision
avoidance systems. The first system initiates decelerations
and is titled automated emergency brake (AEB) system,
whereas the second system initiates a combined deceleration
and evasive maneuver to avoid accidents. This second system
is generally addressed as collision avoidance system (CAS)
within the scope of this paper. The following methodology is
applied to assess the impact of positioning inaccuracies:
1.

2.
3.
4.

1202

Definition of a traffic situation with a vehicle and an


uncertain VRU position and selection of a suitable
measure to assess the safety of this situation
Analysis of the effects of an emergency brake maneuver
executed by vehicle as a reference (AEB system)
Optimization of a combined braking and evading
maneuver (CAS) to reduce the risk in the situation
Assessment of the optimized combined braking and
evading maneuver relative to the brake maneuver by
comparing the resulting risk values

This methodology is applied to a safety critical scenario.


A vehicle driving at the velocity E with a total lateral
width F approaches a VRU crossing the road ahead. The
axes of the vehicle at the start of the situation define the
coordinate system G , H . The VRU crosses the street
exactly at the distance G in front of the vehicle. The
lateral position of the VRU H is assumed to be normally
distributed with the expected value fdemnw and the variance
^
, which results in a probability density function
K
AH for the position of the VRU as given below.


fde

]
,^~



z b}
^
 

 

The collision avoidance system of the vehicle is capable


50 $0/530- 5)& 7&)*$-&@4 #3",&4 "4 8&-- "4 *54 45&&3*/(
actuators. The system influences the lateral position H
as well as the longitudinal position G of the vehicle
after the intervention. Consequently the system also
determines the lateral position H and the velocity
E at the point of the collision with the VRU (see Fig.
3). As a reference also the effects of an emergency braking
system without any lateral intervention is calculated for each
situation.
HE;<

H756

HE;<,<=D

HE;< ,;@:

center of
vehicle

GE;<

right side
of vehicle

GE;< ,<=D

GE;<,;@:

z a

fde fde 

z b



The risk B 0 !" of the traffic situation results from an


evaluation of the severity C caused by the impact velocity and
the hit probability A 0 !" of the VRU.
 xjmmnw 1 xjmmnw 



The severity C 0 !" is defined by the probability of a


fatally injured causality as defined in [12] (see Fig. 1).




xjmmnw !
xjmmnw !
xjmmnw

jyt{a|




]ajyt{a|

 

 #$!*&"(  !!&#$*'see

For the optimization a two step approach is used in the


following. First of all the lateral trajectory is determined and
the longitudinal deceleration profile is added in a second step.
The parameters used to determine the lateral and longitudinal
trajectories are optimized using a genetic algorithm.

The probability AH that the vehicle hits the VRU


is deduced by integrating the probability density function
AH for the total width of the vehicle.
xjmmnw  2

An evasive maneuver requires an accurate planning of the


7&)*$-&@s lateral trajectory considering maximum values for
lateral accelerations. For both lateral and longitudinal
maneuvers tire characteristics are limiting the system
functionalities. In case a combined maneuver applies a
longitudinal deceleration by braking as well as a steering
angle to evade the tire characteristics need to be considered in
more detail. The force transferred by the tire is limited to a
maximum value affecting lateral and longitudinal dynamics
at the same time. Consequently in an evasive maneuver the
system cannot apply the full longitudinal deceleration if
lateral accelerations are required.

while driving [22].

Figure 3. Definition of geometry in the collision scenario

V. OPTIMIZATION OF EVASIVE AND BRAKING MANEUVERS

A. Definition of suitable lateral trajectories


Trajectory planning approaches for robots usually
determine how solutions of motion planning algorithms can
be followed with respect to the mechanical limitations of a
robot [21]. Thus the result of a trajectory planning is a
combination of a path and a feasible velocity profile. An
approach which has been used for decades in robotics is to
compute a collision free path, smooth the path to satisfy some
differential constraints and afterwards design a feasible
velocity function [21]. Because of the non-holonomic
constraints of vehicles the curvature of the path has to be
continuous and smaller than the maximal possible curvature
]
) to perform collision avoidance maneuver
(J 

FE;<

left side
of vehicle

Consequently the risk B is defined in a way that at a value


of B " the VRU is fatally hit by the vehicle. In case the
vehicle is able to decelerate into standstill before hitting the
VRU or a lateral maneuver allows evading from the path of
the VRU the risk value B ! is defined.

The most common model that mathematically considers


this restriction and which is also used in road construction is
the clothoid model in which the curvature changes linearly
with the curve length. In collision avoidance systems
clothoids are sometimes approximated by e.g. third-order
polynominals to decrease computation time [23]. As the
computation of the impact assessment of position
uncertainties is executed offline, the computation time for
this study is less important than the feasibility of the path. For
that reason here the path planning approach shall consider the
non-holonomic constraints of vehicles and a 7th degree
polynomial is used for the evasive path [24] [25].



 

n
xjm xjm ._nc\ n xjm


 

This approach allows considering boundary conditions


for the start and the end position of the evasive path. At the
positions G ! and G G the steering angle of
the vehicle, the steering velocity and the steering acceleration
should be zero. This results in the constraints given in (8-9).



1203

! ! xjmjri xjmjri 

 

iky
iy

i ky
iy

i ky
iy

! h ! xjmjri   

Additionally the lateral acceleration of the trajectory is


constrained by a given maximum value of 8 and
(7-9) are used to determine the coefficients of the
polynomial. The lateral acceleration 8 G results as a
function of the current vehicle speed and acceleration.
x
y

pgw  

ik
iw

i k
iy

iy ^
ik

iw

iy

gy
y
i

iw



th

Contrary to e.g. 5 order polynomials or clothoids, the 7th


order polynomial approach results in a continuous behavior

and thus ensures


of the curvature change

feasibility of the evasive path [24]. The resulting total length
of the evasive path G is longer using the 7th order
polynomial compared to 5th order polynomials, sinus or
clothoid approaches at a given maximum lateral acceleration
8 . Hence in the following the length of the evasive
maneuver is estimated to the safe side.
B. Definition of longitudinal trajectory
In order to depict the forces at the tires of the vehicle
appropriately a friction ellipse connects the longitudinal and
lateral accelerations. The lateral acceleration 8 of the
vehicle may reach a physically defined maximum value of
8 in case the longitudinal acceleration 8 is zero.


g y

g

g y
g



"

This tire model is frequently used in engineering and


maximum values for the accelerations are well known. In the
following the values 8 '?C ^ and 8
)&?C ^ are used to represent average values. The
longitudinal velocity profile is defined based on the
previously defined lateral trajectory. There is no direct
analytic way to combine (10) and (11). To approximate the
velocity profile the lateral trajectory is analyzed beginning at
the distance zero, where the boundary condition E !
E is defined. The velocity profile is defined
recursively. For two distances ! G]  G^ G with
a distance of G  G^  G] the velocities and accelerations
are approximated as given in (12-14).




 pgw ^

ik

 psrl ]  

iy y

i k

 psrl ^  psrlqgy "


xjm ^ xjm ]

^
xjm
]  

iy y

g y
g

gy -y
x y







For G / ! a recursive application of (12) to (14) results


in a trajectory that fulfils the requirement in (11). The
approximation error of G 3 ! can be quantified by
computing the sum in (11). In the following the maximum
error of the approximation is limited to I &+ and hence
only those trajectories are accepted that fulfill (15).



g y

g

g y
g

" 

 

In the calculations G !$? is used in a first step. In


case (15) is violated the distance G is reduced.
C. Optimization using Genetic Algorithms
In order to identify an optimized combined collision
avoidance maneuver utilizing lateral and longitudinal
interventions a genetic algorithm is applied for a specific
traffic situation. Genetic algorithms are generally able to
identify global optima and hence useful for this problem.
Each situation is defined by the distance of the vehicle to
the VRU G , the initial velocity of the vehicle E
and the standard deviation K of the normally distributed
lateral position of the VRU. The algorithm is applied to the
problem to identify optimal values for the parameters

H 0 ! $&? and 8 0 ! *! .

The maximal longitudinal acceleration 8


accepted in a situation determines the polynomial coefficients
(7) and consequently also the total length of the evasive
maneuver G . The two parameters also define the
velocity E and position H when the vehicle hits
the vulnerable road user. This results in a risk B (3) of the
trajectory determined by the variables Hxjmjri and 8 .
The genetic algorithm minimizes a fitness function 4 (16).
0015*.*;& 5)&4:45&.@4 1&3'03."/$& 5)& '*5/&44 '6/$5*0/ 4
is set to a constant value representing the maximum risk 4=2
for all parameter sets that do not result in feasible trajectories.
Otherwise the fitness function is defined by the risk value
(16). The maximum value of the risk of a feasible trajectory
is defined to one. Consequently infeasible trajectories are
punished with a fitness value at least twice as high.
#
 

 



 pgwwugoqgy  xjmjri

 
VI. ANALYSIS OF OPTIMIZATION RESULTS
In the following the effectiveness of a CAS that uses the
optimized trajectory as described in section V and an AEB
system is compared. The methodology defined in section IV
is applied for an urban scenario with a velocity of the vehicle
of E &!  and a VRU crosses the street in
front of the vehicle. Assuming a maximum longitudinal
deceleration of 8 *!?C ^ an AEB system is able
to decelerate the vehicle into standstill within a distance of
G "#*?, which is the safest way to avoid the
collision. The analysis only examines distances G that
do not allow AEB systems or CAS to decelerate into
standstill and a range of %  G "#  is chosen.
The VRU in this scenario crosses the street in different
distances in front of the vehicle and is expected exactly in the
middle of the vehicle (fdemnw = 0 m). The variation of
fdez allows to evaluate the impact of prediction and
detection systems with different accuracies. The evaluation
considers values of !%  fdez # . Smaller values of
fdez are expected e.g. for state of the art camera systems
while higher values could result from satellite positioning
systems with communication between VRU and vehicle. For

1204

each situation an optimization as described in section V.C is


performed in order to identify the CAS maneuver with least
risk for the VRU.
A. Definition of comparison criteria
The potential benefit of a CAS that deploys evasive and
braking maneuvers is compared to the effect of an AEB
system without lateral intervention. As defined above the risk
B 0 !" provides a suitable indicator to assess a trajectory.
Consequently for each situation the risk is deduced for a
braking only maneuver (resulting in the risk B ) and for a
combined braking and evading maneuver (resulting in the
risk B ). In the following the change of the risk B9 by
applying the evasive maneuver relative to the AEB system is
evaluated.




u bu
u

in a reduction of the risk compared to a braking only


intervention of the AEB (Fig. 6). The analysis reveals that the
risk reduction decreases with an increase of K (for
systems with higher prediction inaccuracies) and furthermore
with an decrease of the distance to the VRU G .

 

  0 !"

B. Analysis of risk reduction relative to AEB systems


The following figures visualize the optimization results.
In Fig. 4 the collision velocities using the CAS of the vehicle
is plotted versus the distance to the VRU for different values
of K . As a reference the velocity of the vehicle using the
AEB system is also visualized. The CAS increases the
velocity of the vehicle when hitting the VRU for ) ?
G #  as the system performs evasive maneuvers,
which limit the longitudinal acceleration as reflected by (11).

Figure 5. Probability of the optimized evasive maneuvers xjmmnw of


hitting the VRU for E &!

using the CAS

Figure 6. Change in risk of the CAS (optimized evasive and braking

maneuver) relative to the AEB system (braking only) for E &!

Figure 4. Velocity of the vehicle xjmmnw when hitting the VRU in

comparison to the AEB reference velocity for E &!

In Fig. 5 the probability AH of hitting the VRU is


given for different distances and position accuracies. The
analysis reveals that technologies with K "? are
able to reduce the hit probability especially a G ) .
The resulting change of the risk change B9 of the CAS
compared to an AEB is plotted in Fig. 6 for all values. At
distances of xjmmnw "#  for xjmvwguw &!  the
AEB system is able to reduce the velocity of the vehicle to
zero and consequently the resulting risk value is zero. The
CAS is not able to further reduce the risk in this situation and
also performs a braking maneuver to standstill. For xjmmnw
"#  the CAS is not able to fully avoid a collision but
reduces the collision probability (Fig. 4) while at the same
time increasing the collision velocity (Fig. 5), which results

The standard deviation of the VRU position K


determines the probability of a collision. While Fig. 5
summarizes the collision probabilities for CAS intervention
the collision probability for an AEB intervention without
steering are given in Table I.
TABLE I.
ASSESSSMENT OF HIT PROBABILITY FOR AEB
INTERVENTION (BRAKING ONLY) COMPARED TO HIT POBABILITY OF CAS
INTERVETION (COMBINDED STEERING AND BRAKING)
Standard deviation
of VRU position
YQOPX

Hit probability of AEB


without steering
intervention ULNM

Minimum hit
probability of CAS at
WVRS Z[T

K #!?

A H $&(+

A H $"#+

K "!?

A H '%'+

A H #"$+

K !(?

A H )"&+

A H "#!+

K !&&?

A H *!)+

A H '&+

K !%?

A H *)!+

A H ")+

1205

The activation of an emergency braking or evasive


steering maneuver is highly critical and false activations need
to be strictly limited. The comparison in Table I reveals that
the collision probability A exceeds 90 % only atK
!&&? and consequently for these values low rates of false
activation are expected. For E &!>?< the
analysis thus reveals an acceptable standard deviation of the
VRU position of K !&&? ensuring a false
activation ratio of less than 10 %. Here the CAS offers a
significant risk reduction potential compared to the AEB
system. The optimized lateral and longitudinal trajectories are
visualized in Fig. 7. The optimizer identifies a minimum risk
for a combined evasive and deceleration maneuver with a
lateral offset of about 1 m and a velocity of around 30 km/h
at the VRU. The computed lateral and longitudinal
accelerations fulfill (15) with I "!b` . The vehicle
decelerates to standstill within 15 m although the evasive
maneuver is planned up to a distance of 22 meters.

[3]  *48"4  "5$)*,06  *0/ = &)*$-&-to-Vehicle Wireless


[4]
[5]
[6]
[7]
[8]

[9]

[10]

[11]
[12]

[13]
[14]
[15]
[16]
[17]
Figure 7. Optimal evasive and braking trajectory at xjmvwguw &! 
for K !&& and G "!?

[18]

VII. CONCLUSION AND OUTLOOK

[19]

)*4803,13&4&/54"/"1130"$)50015*.*;&5)&7&)*$-&@4
longitudinal and lateral trajectories in critical situations by
minimizing the risk for VRU. A method to approximate
combined braking and evading trajectories is proposed and
applied to analyze the impact of positioning and prediction
inaccuracies of VRU on the risk. The analysis of a typical
situation reveals that positioning accuracies of less than
K !&& ? are required in order to significantly
reduce the risk of VRUs by deploying evading and braking
instead of braking only emergency maneuvers.
REFERENCES
[1] !03-%&"-5)3("/*;"5*0/=&%&453*"/4"'&5:  30"% safety manual

[20]

[21]
[22]
[23]

[24]

for decision-.",&34"/%13"$5*5*0/&34?

[2] E. E. Nodine, A. H. Lam, J.  &3&/$& ="'&5: .1"$5 0' "/


/5&(3"5&% 3"4) !"3/*/( :45&. "4&% 0/ *&-% &45 "5"? nd
ESV, 2011

[25]

1206

0..6/*$"5*0/ 3050$0-4 '03 /)"/$*/( *()8": 3"''*$ "'&5:?


IEEE Communications Magazine, 2006
&33"3"= 650."5*$3&-3"4)0--*4*0/ 70*%"/$&*/"34?
Intelligent Vehicles Symposium, 2004
"#":3"%&0:&3& 6#&35= 0--*4*0/*tigation System
64*/( "4&3 $"//&3 "/% 5&3&07*4*0/ 64*0/ "/% *54 44&44.&/5?
IEEE Intelligent Vehicles Symposium, 2005
M. Ahrholdt, P. Pyyknen, M. Kutila, J. Scholliers, J. Ylihao, J.
!&*/("35  &  0&44-&3 =/5&34&$5*0/ "'&5: :45&.
Demonstra503=/5&34"'&&-*7&3"#-&
%&
 "/%)*   3*7&%* =&%&453*"/ 305&$5*0/ :45&.4  446&4
637&: "/% )"--&/(&4?  3"/4"$5*0/4 0/ /5&--*(&/5
Transportation Systems, Vol. 8, 2007
G. R. de Campos, A. H. Runarsson, F. Granum, P. Falcone, K.
-&/-+6/( =0--*4*0/ "70*%"/$& "5 */5&34&$5*0/4   130#"#*-*45*$
threat-assessment and decision-making system for safety
*/5&37&/5*0/4?  
th International Conference on Intelligent
Transportation Systems, 2014
K. Ismail, T. Sayed, N. Saunier"/%*.= 650."5&% /"-:4*40'
Pedestrian- &)*$-& 0/'-*$54 4*/( *%&0 "5"? Transportation
Research Record: Journal of the Transportation Research Board, vol.
2140, no. -1, pp. 44<54, 2009
D. Fernndez Llorce, V. Milans, I. P. Alonso, M. Gaviln, I. G.
";"  B 05&-0 = 650/0.064 &%&453*"/ 0--*4*0/ 70*%"/$&
4*/("6;;:5&&3*/(0/530--&3?3"/4"$5*0/40//5&--*(&/5
Transportation Systems, Vol. 12, 2011
ISO 26262- =0"% 7&)*$-&4 < Functional safety < Part 1:
0$"#6-"3:?
M. %8"3%4  "5)"/40/  !*4$) = &/&'*5 &45*."5& "/%
assessment methodologies for pre-crash braking part of forward-00,*/( */5&(3"5&% 1&%&453*"/ 4"'&5: 4:45&.4? 4& &-*7&3"#-&
1.3, 2014
J. J. Anaya, P. Merdrignac, O. Shagdar, F. Nashashibi, and J. E.
"3"/+0 = &)*$-& 50 &%&453*"/ 0..6/*$"5*0/4 '03 305&$5*0/ 0'
6-/&3"#-&30"%4&34?*/Intelligent Vehicles Symposium, 2014
! &3)"3%  $).*5; "/%  "((& =-RTK: Precise Point
Positioning using state-space representation in RTK networks,?
Proceedings of ION GNSS, 2005, 2005
L. Wang, P. D. Groves, and M. Ziebart, Urban Positioning on a
Smartphone: Real-time Shadow Matching Using GNSS and 3D City
Models.
H. Dodel and D. Hupler, Eds, Satellitennavigation. Berlin,
Heidelberg: Springer Berlin Heidelberg, 2010
B. Hofmann-Wellenhof, H. Lichtenegger, and E. Wasle, GNSS-global navigation satellite systems: GPS, GLONASS, Galileo, and
more. Wien, New York: Springer, 2008
  +6,/*$ "/%   *$)50/ =&0-0$"5*0/ "/% "44*45&% ?
Computer, vol. 34, no. 3, pp. 123<125, 2001
S. Khler, M. Goldhammer, S. Bauer, S. Zecha, K. Doll, U.
Brunsmann, K. Dietmayer, =5"5*0/"3: &5&$5*0/ 0' 5)& &%&453*"/@4
/5&/5*0/ "5 /5&34&$5*0/4? /5&--*(&/5 3"/41035"5*0/ :45&.4
Magazine, Vol. 5, Issue 4, 2013
D. Westhofen, C. Grundler, K. Doll, U. Brunsmann, and S. ZECHA,
=3"/410/%&3- and Camera-#"4&%"%7"/$&%%3*7&3"44*45"/$&4:45&.?
in Intelligent Vehicles Symposium (IV), 2012 IEEE, 2012, pp. 293<
298
S. La Valle, Planning Algorithm, Cambridge University Press,
Cambridge, 2006
3"*$)"3%"/% $)&6&3=30.&&%4"/%)&11450$0/5*/6064$637"563& 1"5)4? IEEE TRANSACTIONS ON ROBOTICS AND
AUTOMATION, vol. 20, no. 6, pp. 1025<1035, 2004
A. Polychronopoulos, M. Tsogas, A. J. Amditis, and L. Andreone,
=&/40364*0/'033&%*$5*/( &)*$-&4"5)'030--*4*0/ 70*%"/$&
:45&.4?Intelligent Transportation Systems, IEEE Transactions on,
vol. 8, no. 3, pp. 549<562, 2007
C. Keller, T. Dang, H. Fritz, A. Joos, C.Rabe, D. M. Gravila, =Active
&%&453*"/"'&5:#: 650."5*$ 3",*/("/%7"4*7&5&&3*/(=IEEE
Transactions on Intelligent Transportation Systems, vol. 12, no 4, pp.
1292-1304, 2011
 -&%(&  "34)&, =0.1"3*40/ 0' %&"- &)*$-& "/&-Change
3"+&$503*&4? &$)/*$"-"1&3971062, 1997

Vous aimerez peut-être aussi