1802 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO.

4, MAY 2014

A Vehicular Networking Perspective on Estimating

Vehicle Collision Probability at Intersections
Stefan Joerer, Student Member, IEEE, Michele Segata, Student Member, IEEE,
Bastian Bloessl, Student Member, IEEE, Renato Lo Cigno, Senior Member, IEEE,
Christoph Sommer, Member, IEEE, and Falko Dressler, Senior Member, IEEE

Abstract—Finding viable metrics to assess the effectiveness of but the first application type, i.e., safety, adds further demands
intelligent transportation systems (ITSs) in terms of safety is one such as extremely low transmission latencies combined with
of the major challenges in vehicular networking research. We high communication reliability [6], [7].
aim to provide a metric, i.e., an estimation of the vehicle colli-
sion probability at intersections, that can be used for evaluating With the development and standardization of dedicated short-
intervehicle communication (IVC) concepts. In the last years, the range communication (DSRC) using IEEE 802.11p at the ac-
vehicular networking community reported in several studies that cess level [8], short-range radio broadcast became a viable
safety-enhancing protocols and applications cannot be evaluated complement to cellular communications and one of the pre-
based only on networking metrics such as delays and packet loss ferred technologies for low-latency communications between
rates. We present an evaluation scheme that addresses this need
by quantifying the probability of a future crash, depending on vehicles in close vicinity. DSRC promises to reduce accidents
the situation in which a vehicle is receiving a beacon message by enabling novel support systems. Within this scope, a wide
[e.g., a cooperative awareness message (CAM) or a basic safety range of applications has been identified, from emergency
message (BSM)]. Thus, our criticality metric also allows for fully braking systems for highways [9] to more radical innovations
distributed situation assessment. We investigate the impact of such as virtual traffic lights [10], [11].
safety messaging between cars approaching an intersection using
a modified road traffic simulator that allows selected vehicles to One of these applications is intersection collision warning
disregard traffic rules. As a direct result, we show that simple bea- systems (ICWSs), which can offer real-time warnings up to
coning is not as effective as anticipated in suburban environments. fully automated reactions [12]–[14]. The benefit of such sys-
More profoundly, however, our simulation results reveal more tems has already been investigated thoroughly using driving
details about the timeliness (regarding the criticality assessment) simulators. In 2009, Chang et al. have shown that audio-based
of beacon messages, and as such, they can be used to develop more
sophisticated beaconing solutions. ICWSs are able to reduce drivers’ reaction time and hence
reduce the accident rate [12], e.g., for young inexperienced
Index Terms—Vehicle safety, vehicular ad hoc networks, wire- drivers. The impact of different warning systems has been
less communication.
studied in [15], and each investigated type clearly indicates
a substantial safety advantage. These early results show the
I. I NTRODUCTION
potential of ICWSs as the number of intersection crashes could

C RASH mitigation and crash avoidance are two of the

major applications of intelligent transportation systems
(ITSs) [1]. Most recent approaches for active safety also take
be reduced by 40% to 50%. However, these results should be
seen as a baseline as they neither address how ICWS can be
implemented nor consider the involved networking issues.
intervehicle communications (IVC) into consideration [2]. In In this paper, we employ simple beaconing, i.e., one-
general, research on IVC is mainly motivated by two classes hop broadcasts, for exchanging safety critical information via
of applications: safety and efficiency. Both require proper man- DSRC in the context of ICWSs at suburban intersections.
agement of the wireless communication channel [1], [3]–[5], Third- and fourth-generation approaches are also considered for
this application scenario [16], [17] but are outside the scope
of this paper. Beaconing has been identified in the literature
Manuscript received April 15, 2013; revised August 23, 2013; accepted as a communication strategy suitable for many challenging
October 15, 2013. Date of publication October 25, 2013; date of current version vehicular networking applications [3], [9], [18], [19]. It is being
May 8, 2014. The review of this paper was coordinated by Prof. J. Wang.
S. Joerer, B. Bloessl, C. Sommer, and F. Dressler are with the Computer and
standardized for the dissemination of safety critical information
Communication Systems Group, Institute of Computer Science, University of to be broadcast periodically at 1 to 25 Hz in the form of
Innsbruck, Innsbruck 6020, Austria (e-mail: [email protected]; bloessl@ccs- cooperative awareness messages (CAMs) [4] and basic safety
labs.org; [email protected]; [email protected]).
M. Segata is with the Institute of Computer Science, University of Inns- messages (BSMs) [20] in Europe and in the U.S., respectively.
bruck, Innsbruck 6020, Austria, and also with the Department of Information In previous work [7], we also explored possible improvements
Engineering and Computer Science, University of Trento, Trento 38123, Italy using available relay nodes (e.g., parked vehicles [21] close to
(e-mail: [email protected]).
R. Lo Cigno is with the Department of Information Engineering and Com- the intersection). In this paper, we focus on the possibility of
puter Science, University of Trento, 38123 Trento, Italy (e-mail: locigno@ estimating the collision probability when a beacon message is
disi.unitn.it). received. The approach is fully distributed and does not require
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. external infrastructure. Our goal is to gain insight into the
Digital Object Identifier 10.1109/TVT.2013.2287343 requirements of the communication platform, but (unlike earlier

0018-9545 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 foratmore information.
07:06:20 UTC from IEEE Xplore. Restrictions apply.
 JOERER et al.: VEHICULAR NETWORKING PERSPECTIVE ON ESTIMATING VEHICLE COLLISION PROBABILITY AT INTERSECTIONS 1803

work) we approach this problem from the wireless networking [4], [20]. In the vehicular networking community, approaches
perspective. clearly outperforming simple beaconing in terms of channel
Two issues have routinely been overlooked in the past. First, load or information dissemination range have been proposed.
networking-related metrics often do not reveal the quality of DV-Cast [23] aims at mitigating the broadcast storm problem
ITS-based solutions [2]. Second, the gap between application by rebroadcasting first (and hopefully only) from the vehicles
requirements and networking concepts needs to be closed [22]. farthest from the original sender. The protocol can also switch
In particular, in most vehicular ad hoc network studies on between relaying and opportunistic forwarding depending on
safety and safety applications, the performance of the appli- the estimated advantages. An initial work on adaptive beacon-
cations was not measured through safety metrics, although the ing is Adaptive Traffic Beacon [3], which continuously adapts
final goal of these applications is to investigate the benefit that to the available channel capacity by modifying the beaconing
they are able to provide for the driver and not delay or packet interval. Beaconing and adaptive changes of the beaconing
loss. Therefore, we believe that it is important that future pro- interval have also been investigated in many other publications
posals are not analyzed with network metrics such as latency, [18], [19].
goodput, or dissemination area, but that studies concentrate Based on these studies, decentralized congestion control
on safety metrics [2], answering more relevant questions such (DCC) has been suggested in ETSI ITS-G5 to cope with
as how many crashes can be avoided, and can the impact of congestion problems [24], [25], and more advanced dynamic
crashes be significantly reduced? Accordingly, we develop new beaconing approaches have been proposed [5]. Nevertheless,
safety metrics and show in this paper, which extends earlier optimizations in this network-specific domain are not the focus
work presented in [7], how these reflect the performance of of this paper.
simple beaconing based communications. When looking at the communication perspective of inter-
Our main contributions can be summarized as follows. section applications, most approaches did not evaluate their
• Collision probability estimation. We built upon the ini- communication systems using safety metrics [10], [11], [26].
tial coarse-risk classification that we presented in [7] to Le et al. looked at the busy time fraction of DSRC systems
develop a more comprehensive estimation of criticality, for intersection safety [26] using a simplified radio propagation
which is now expressed as collision probability. This model that uses only a fixed unit-disk communication range.
probability is a quantitative measure of the criticality of A detailed study on communication requirements for crash
intersection approach situations (see Section III). avoidance applications has been published in [27]. The authors
• Integration into a road traffic simulator. We developed changed collision-free vehicle traces by artificially injecting
a simulation environment that enables the (collision-free) collisions with constant velocity to evaluate their protocol in
road traffic simulator SUMO to support vehicles that se- terms of crash mitigation possibilities. However, simplifying
lectively ignore traffic rules; we further integrated the pos- assumptions such as idealistic radio signal propagation and
sibility of detecting the resulting crashes or near crashes not considering low-speed collisions (< 7 m/s) limit the con-
(see Section IV). tribution for intersection safety applications. We go one step
• Validation of the probability estimation. We carefully eval- further and evaluate the ICWS with new safety metrics that
uated our collision probability using measurements from a are based on the collision probability of two approaching
high number of simulated intersection approaches, using vehicles.
lane geometry imported from OpenStreetMap (presented Tang and Yip [28] investigated timings for collision avoid-
in Section V-A). Using two simple example models, we ance systems assuming DSRC transmission delays of 25 and
show how our mathematical approach can be adapted to 300 ms in normal and poorer channel conditions, respectively.
capture different driver behaviors (see Section V-B and C). They introduced the time-to-avoid collision metric, which rep-
• Implications on vehicular networking concepts: We study resents the time from detecting a potential collision to the point
the impact of beaconing for the transmission of safety of barely avoiding a collision and concentrated on the events
messages in non-line-of-sight scenarios, investigating the (when to warn a driver early and latest, reaction of the driver,
timeliness (regarding the criticality assessment) of beacon and different deceleration rates) within this time interval. This
messages. Our results indicate that simple beaconing is metric is definitely a good possibility for comparing informa-
not as effective as anticipated in suburban environments; tion dissemination protocols for intersection safety; however,
further insights can be used to develop more sophisticated the point in time for detecting a potential collision has not
CAM/BSM beaconing solutions (see Section V-D). been defined explicitly, and their analysis is limited to two fixed
transmission delays, which in reality will vary.
Research that focuses on estimating, predicting, and/or re-
II. R ELATED W ORK
ducing the likelihood of crashes at an intersection provides
This paper focuses on collision avoidance at intersections as various approaches to model intersection approaching vehi-
one application of ITS; hence, it touches on not only communi- cles. This research goes as far as to include threat assessment
cations issues but also on research areas such as control theory, for avoiding arbitrary collisions with bicycles. For this, lat-
transportation science, and road traffic engineering. eral and longitudinal movements and vehicle dynamics have
Starting with the communication perspective, we investigate been modeled [29], yet communication aspects have not been
the possibilities of intersection safety applications using sim- investigated. Lefèvre et al. [30] point out that risk assess-
ple beaconing strategies as currently proposed in standards ment at intersections is possible by comparing intention and
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 at 07:06:20 UTC from IEEE Xplore. Restrictions apply.
 1804 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 4, MAY 2014

expectation. Liebner et al. [31] also used drivers’ intent infer-

ence at intersections.
In our opinion, an approach that restricts the analysis of
communication protocols to one particular estimated driver
behavior is not applicable. Therefore, we decided to use an-
other approach to modeling safety aspects for intersection-
approaching vehicles: model the probability of all possible
future trajectories, and exploit their likelihood to estimate Fig. 1. Coordinate space for vehicles A and B for different intersection types.
collision probabilities. In this area, Tan and Huang [32] have (a) X-intersection. (b) Y-intersection.
explored the possibilities of future trajectory prediction for
cooperative collision warning systems with a focus on the
engineering feasibility using simple GPS receivers and motion
sensors.
Verma and Del Vecchio [33] presented a hybrid control
approach for cooperative active safety systems, which was
evaluated in the laboratory using robots. However, even under
laboratory conditions, it becomes clear that the communication
delay is critical for the controller and even caused failures.
Hafner et al. [34] have built an automated vehicle-to-vehicle
(V2V) collision avoidance application that avoids collisions by
automatically controlling the longitudinal movements of both
Fig. 2. Sample trajectory of vehicle A dependent on its distance dA and
vehicles. The decision whether the system needs to control the speed vA . Distances denter and dleave , and times tenter and tleave are shown
brake and throttle is based on the calculation of the capture set, for an orthogonal X-intersection.
which is the set of all situations where no control input is able
to prevent a collision [35] and the prevention of such situations. but depend on each vehicle model. To simplify the notation in
They showed that the controller is able to avoid collisions this paper, we omit the vehicle-dependent indexes for these two
under favorable communication conditions with a two-car real physical boundaries in the following.
testbed. For defining distances dA and dB , the intersection is modeled
In contrast, this paper focuses on the communication aspects; as a simple coordinate space, where the axes are defined by
hence, its goal is not to design a novel vehicle controller. Thus, the future driving path of the vehicles and are not necessarily
we abstract the aspects (i.e., lateral and longitudinal vehicle orthogonal (cf. Fig. 1). The axes’ origins are at the center of
dynamics, sensor errors, and feedback control) that would where the vehicles’ trajectories intersect. In the following, we
be needed in this case, focusing instead on the identification concentrate on the X-intersection shown in Fig. 1(a). However,
of communication conditions that would hamper any control by considering the interdependence of the two distances, a
system. Y-intersection, as shown in Fig. 1(b), can be modeled similarly.
Our approach uses a probabilistic model for trajectories to
represent all possible future driver behaviors and to derive A. Trajectories
collision probabilities, but we do not attempt to take decisions
that influence their future evolution (triggering an automated To define the intersection collision probability, we first need
reaction or warning a driver), leaving these aspects for future to mathematically model all possible driver behaviors of a
works. We use the presented intersection collision probability to single vehicle.
evaluate the communication aspects of ICWSs, and it provides Depending on the current distance dA and speed vA of
the potential for enhancing future communication strategies for vehicle A, an unlimited number of future trajectories TA (i.e.,
intersection safety applications. different intersection approaches) are possible. With current
time being t0 , a trajectory is a feasible function of time that
describes the vehicle’s distance from the intersection center
III. I NTERSECTION C OLLISION P ROBABILITY respecting the initial conditions and acceleration limits, i.e.,
Our goal in establishing a criticality metric is to calculate
the probability of a possible collision whenever we have new TA (t0 ) = dA ṪA (t0 ) = vA , amin ≤ T̈A (t) ≤ amax . (1)
information about two potentially colliding vehicles available, Given the current distance dA and speed vA of vehicle A,
i.e., every time a car receives a beacon message (which includes we callthe measurable set of all possible future trajectories
position information speed, heading, etc., of the sender). In our TA = TA . This set is limited by the maximum acceleration
case, the needed information for two approaching vehicles A amax and maximum deceleration amin , as shown in Fig. 2.
and B consists of the distances from their trajectories’ inter- To determine whether a collision happens for two trajectories
section dA and dB and the speeds vA and vB , as well as the TA ∈ TA and TB ∈ TB , we define the function coll (TA , TB ) as
maximum acceleration amax and the maximum deceleration (in 
terms of a minimum, negative acceleration) amin . Notably, amin 1, if there is a collision
coll (TA , TB ) = (2)
and amax are not necessarily the same for vehicles A and B 0, otherwise
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 at 07:06:20 UTC from IEEE Xplore. Restrictions apply.
 JOERER et al.: VEHICULAR NETWORKING PERSPECTIVE ON ESTIMATING VEHICLE COLLISION PROBABILITY AT INTERSECTIONS 1805

where we define a collision as occuring if the bounding boxes

of the vehicles are overlapping at some point in time during the
intersection approach.

B. Definition of Collision Probability Fig. 3. Example of a triangular acceleration probability distribution con-
ditioned on the present acceleration (solid line) compared with a uniform
If we integrate over all possible trajectories TA and TB of distribution (dashed line).
two approaching vehicles, we can define the probability PC of
a collision at an intersection as straint, every trajectory T can be identified by a tuple (a, v, d),
  and we can define a new function coll(·, ·) analogous to (2) but
PC = p(TA , TB ) coll (TA , TB ) dTA dTB . (3) only depending on these values. Hence, we can calculate PC by
TB TA
integrating over the interval amin and amax for both vehicles as
follows:
The function p(TA , TB ) gives the probability that the trajecto- amax amax
⎛⎡ ⎤ ⎡ ⎤⎞
aA aB
ries TA and TB are chosen and hence provides the possibility of
PC = p(aB ) p(aA ) coll ⎝⎣ vA ⎦ , ⎣ vB ⎦⎠ daA daB .
modeling different kinds of driver behavior. In particular, this
amin amin dA dB
general definition of the collision probability does not assume
(4)
the two chosen trajectories to be independent of each other.
Moreover, our calculated collision probability does not distin- The behavior of drivers, i.e., how likely it is that a driver
guish situations where a crash has happened already (which is chooses a certain acceleration, can now be modeled by defining
called a bad set in [34]) and a future crash is unavoidable (which the distribution of accelerations. In the following, we present
is called a capture set in [34]); PC will in both situations be two possible simple distributions to give an idea of their impact
100%. In the following, we continue with a simplified version on collision probability.
of this general approach because, to evaluate communication
strategies for ICWSs, we do not need to model details such
D. Uniform Acceleration Probability Distribution
as lateral movements and/or longitudinal vehicle dynamics, for
example. One simple example is a uniform distribution of all possible
accelerations between amin and amax . We will use this distribu-
tion to demonstrate the applicability of the collision probability
C. General Assumptions
defined in (3). Probability p(a) can be then calculated as
The formulation presented is very general and has high 
1
expressive power. However, without some additional assump- p(a) = amax −amin , if amin ≤ a ≤ amax (5)
tions, it is hardly tractable. Thus, we now introduce several 0, otherwise
simplifying assumptions that can be selectively relaxed when resulting in
additional insight on a specific issue is needed. As a first ⎛⎡ ⎤ ⎡ ⎤⎞
amax amax
simplification, in the following, we consider only orthogonal aA aB
1
X-intersection crossings without turning maneuvers. In this PC = coll ⎝⎣ vA ⎦,⎣ vB ⎦⎠ daA daB .
(amax −amin )2
case, a collision happens for two given trajectories if both amin amin dA dB
vehicles are in the potential collision area, i.e., where the ve- (6)
hicles might hit/touch each other [shown in Fig. 1(a) as orange
crosshatched area] at the same time. The size of the potential
E. Towards More Realistic Driver Behavior
collision area depends only on the vehicles widths. Thus, the
times tenter and tleave , i.e., when a vehicle enters and leaves As a uniform acceleration distribution does not account for
the potential collision area of a given trajectory, respectively, the current acceleration of the car, it might not represent typical
can be calculated using the trajectory and the distances denter human driver behavior well. It might be considered more likely
and dleave . The relationship between a sample trajectory TA , that the driver continues to drive with the current acceleration;
the times tenter and tleave , and the distances denter and dleave is similarly, extreme accelerations could be very unlikely. One
shown in Fig. 2. possibility for representing such behavior is to employ a tri-
As a second simplification, we assume that the probabilities angular acceleration probability distribution with lower limit
for the two trajectories TA and TB are independent. Currently, amin , mode acur , and upper limit amax , as shown in Fig. 3.
the literature does not give insight into whether and to what When using this distribution, the collision probability PC can
degree two approaching vehicles might influence the behavior still be calculated using (4).
of each other (causing a driver to accelerate, decelerate, or
swerve). Moreover, we are particularly interested in situations
IV. S IMULATION M ODEL AND S ETUP
where the drivers are not aware of each other; hence, the
probability of choosing a certain trajectory does not depend on We conducted an extensive simulation study to validate
the other one. Furthermore, we consider only trajectories with a and evaluate the proposed collision probability estimation. For
constant acceleration between amin and amax . Under this con- this, we used version 2.0 of the vehicular network simulator
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 at 07:06:20 UTC from IEEE Xplore. Restrictions apply.
 1806 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 4, MAY 2014

TABLE I TABLE II
ROAD T RAFFIC S IMULATION PARAMETERS , I NCLUDING C OMMUNICATION S IMULATION PARAMETERS FOR S IGNAL
C AR -F OLLOWING PARAMETERS FOR IDM ATTENUATION , P HYSICAL L AYER , AND MAC

larity for the vehicle movements. Considering the maximum

possible speed vmax and the maximum deceleration and accel-
eration amin and amax , respectively, of the vehicle, we can com-
pute the maximum possible error  introduced by deceleration
and acceleration as
1 1
εdec = |amin |t2step , εacc = |amax |t2step (7)
2 2
ε = max(εdec , εacc ). (8)

In our simulations, we choose time step tstep = 5 ms for

Fig. 4. Map view of the simulated X-intersection showing the potential simulating the longitudinal movements within SUMO, which
collision area, buildings, and two approaching vehicles.
leads to a maximum error  = 0.12 mm (i.e., the vehicle is
braking with maximum deceleration). Moreover, the selected
Veins [36], which bidirectionally couples the road traffic simu- simulation time step results in a maximum step distance of
lator SUMO and the network simulator OMNeT++. It extends vmax × tstep = 8.405 cm and allows us to detect vehicle col-
the MiXiM physical-layer simulation framework and provides lisions very accurately. Still, with a low probability, we might
a rich set of simulation models for realistic simulation of IVC fail to detect slightly “touching” vehicles; how we account for
protocols and applications. these inaccuracies is detailed in Section V.

A. Modeling Road Traffic and Crash Situations B. Modeling Communication

We simulated a typical suburban X-intersection, which is All communication-relevant parameters are summarized in
based on lane geometry imported from OpenStreetMap, and let Table II to ensure reproducibility and comparability for future
two vehicles approach the intersection, then cross it without simulation studies. For the radio propagation, we use the ob-
turning. For this paper, we used the intelligent driver model stacle model [39] that allows us to accurately model signal
(IDM), the car-following model [37] (to reproduce realistic attenuation by buildings in a computationally efficient way.
braking behavior [13]), and a modified version of SUMO For simulating the physical and medium access control (MAC)
that allows us to let selected vehicles ignore traffic rules [7]. layers, we make use of the well-validated IEEE 802.11p model
We randomly selected 50% of approaching vehicles to ignore [40], configuring it to represent a single-radio/single-channel
traffic rules. For inducing situations of different criticality at DSRC system with parameters as listed in Table II.
the intersection, the two vehicles used random initial speeds, The exchange of CAMs/BSMs is implemented as a static
maximum speeds, and desired deceleration values as listed in beaconing application that generates a beacon containing the
Table I. The variation of IDM parameters resembles different needed information for calculating the collision probability
driver behaviors. Since we want to evaluate intersection warn- every beacon interval and passes the message down to the
ing applications regarding their communication requirements, MAC. We simulate four different beacon intervals (every 0.04,
our simulated intersection approaches do not resemble lateral 0.1, 0.5, and 1 s) in the range of the DCC defined in ETSI
and longitudinal vehicle dynamics. standard [25],
In Fig. 4, the potential collision area of two approaching
vehicles, the intersection, and vehicle geometry are shown
V. E VALUATION
in detail, Since vehicles in this simulation study cross the
intersection without turning, this yields a rectangular potential The presented results are based on a simulation study, as
collision area of size 1.75 m × 1.75 m. described in Section IV, and we want to emphasize that the
Since our simulation study is based on the discrete road vehicle movements are controlled by IDM with varying input
traffic simulator SUMO and the linearly interpolating mobility parameters and hence result in different driver behaviors and
model of Veins, we assessed the necessary simulation granu- a wide variety of intersection approaches. In the following,
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 at 07:06:20 UTC from IEEE Xplore. Restrictions apply.
 JOERER et al.: VEHICULAR NETWORKING PERSPECTIVE ON ESTIMATING VEHICLE COLLISION PROBABILITY AT INTERSECTIONS 1807

we investigate the intersection collision probability estimation

based on three different outcomes of intersection approaches;
we distinguish the outcomes as follows. The first group C RASH
includes only intersection approaches of vehicles that collided
at the intersection. To be able to distinguish critical and non-
critical situations better, we introduce a second group called
N EAR C RASH that includes all approaches where a vehicle’s
safety boundary of 0.4 m has been violated by the other one.
This group covers also crash situations that have not been
detected due to simulation time step size and allows for the
detection of situations in which a driver would already feel quite
unsafe. The third group N O C RASH contains only intersection
approaches where the vehicles did not collide nor violate the
safety boundary of each other. Fig. 5. Boxplot showing the maximum collision probability per approaching
vehicle calculated using exact sensor data and grouped by the outcome.
For each simulation parameter set, we simulate 5000 inter-
section approaches using the parameters in Table I and record
the successfully received beacons, the exact movements of the geodata) and the perfect intersection layout assumed during
approaching vehicles, and the outcome of each approach at the probability calculation. To cross-check, we simulated a perfect
intersection. The distribution of all simulation runs across these intersection with the exact (down to the millimeter) layout
groups has been as follows: 3.76% of runs resulted in C RASH, shown in Fig. 4. In this perfect intersection scenario, every
0.84% N EAR C RASH, and 95.4% N O C RASH. approaching vehicle reached a maximum collision probability
of 100% (data not shown).
Finally, we can observe that, in group N EAR C RASH, prob-
A. Validation of Collision Probability
abilities are at a median value of 42%, with some interesting
As a first step, we investigate whether the defined collision outliers at 100%. We analyzed the outliers and found that all
probability is behaving as intended for both the most general of them depict vehicle collisions that went undetected due to
form of acceleration probability distribution (uniform) and the simulation time step size.
more realistic distribution (triangular). The collision probability We can conclude that, under the given assumptions and with
should have the following two properties essential for vehicular perfect knowledge, the algorithm shows no false positives and
safety application. no false negatives.
• No false positives. During N O C RASH approaches, the
collision probability estimation should never exceed a
certain threshold. B. Bringing Networking Into the Picture
• No false negative. During C RASH approaches, the colli- We now go one step further and calculate the collision
sion probability estimation should exceed at least a certain probability based on the CAM/BSM information received from
threshold (ideally close to 100%). the other vehicle. The calculation is carried out using exact local
To validate the collision probability, we recorded the exact information together with delayed CAM/BSM data received
position, speed, and acceleration for each time step and vehicle from the other vehicle. Please note that our scenario with
without considering any communication delay (referred to as only two approaching vehicles periodically sending CAM/BSM
sensor data). Based on these data, we calculated the maximum messages represents an ideal (best) case with respect to channel
collision probability for each approaching vehicle and present conditions and medium access delay.
the distribution in box plots grouped by the different outcomes Fig. 6 shows the distribution of the maximum collision prob-
in Fig. 5. For each data set, a box is drawn from the first quartile ability per vehicle for different beacon intervals (0.04, 0.1, 0.5,
to the third quartile, and the median is marked with a thick line; and 1 s), which are grouped by the three different outcomes of
additional whiskers extend from the edges of the box toward an intersection approach. Basically, we can observe that, for the
the minimum and maximum of the data set but not further group N O C RASH , the distribution of maximum collision prob-
than 1.5 times the interquartile range. Data points outside the ability does not significantly change for the different beacon
range of box and whiskers are considered “outliers” and drawn intervals. Hence, we can conclude that the beacon frequency
singularly as circles. has no effect on false positives when using a warning threshold
When looking at the N O C RASH group, we see that the of 40%. However, for the group C RASH , we see that the number
median probability value is approximately 10%, and even the of vehicles that do not reach a high collision probability is sub-
highest value is clearly smaller than 40%. stantially increasing for larger beacon intervals. This is shown
For the C RASH group, we see that almost all approaching ve- by comparing Fig. 6(a), which shows a similar distribution as
hicles have reached a maximum collision probability of 100%. shown in the validation (cf. Fig. 5), and Fig. 6(d), which reveals
We carefully checked all the vehicles for which a percentage that already a major portion of the approaching vehicles has not
smaller than 100% has been recorded and observed that these reached a high collision probability before crashing.
false negatives occur due to a mismatch of the simulated To understand the correlation between the vehicles’ distances
intersection layout (which is based on the mentioned real-world and the resulting estimated collision probabilities, Fig. 7 shows
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 at 07:06:20 UTC from IEEE Xplore. Restrictions apply.
 1808 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 4, MAY 2014

Fig. 6. Comparison of the maximum collision probability per approaching vehicle for all received beacons, which are grouped by the final outcome of the
intersection approach. The results for different beacon intervals are shown. (a) Beacon interval of 0.04 s. (b) Beacon interval of 0.1 s. (c) Beacon interval of 0.5 s.
(d) Beacon interval of 1 s.

Fig. 7. Mean estimated collision probability per bin calculated based on beacons received using a beacon interval of 0.04 s and assuming uniform distribution
of possible trajectories; the dot size represents also the collision probability of the bin by showing large dots for high collision probabilities. (a) C RASH.
(b) N EAR C RASH. (c) N O C RASH.

the mean collision probability that has been reached for the
vehicles’ positions at the time a beacon was received. We
binned all received beacons by the distance of the sender and
the receiver to the intersection; the mean collision probability is
calculated for each of the resulting bins and is depicted by the
color and the size of the dots in the plot. The presented plots
are separated for the three outcomes C RASH, N EAR C RASH,
and N O C RASH, and show the calculated collision probabilities
for a beacon interval of 0.04 s.
Let us first concentrate on Fig. 7(a), where the estimated
mean collision probabilities of only C RASH approaches are
plotted. As expected, all points close to the potential collision
area show a very high mean collision probability, which is
steadily decreasing when looking at points farther away from Fig. 8. Maximum collision probability per approaching vehicle for all re-
the intersection. For outcome N EAR C RASH [cf. Fig. 7(b)], ceived beacons (beacon interval 1 s) using the triangular distribution.
no beacons have been received close to the diagonal, and the
C. More Realistic Driver Behavior
estimated collision probabilities reach a medium level (about
50%) at a distance of 20 m, but they decrease again toward As mentioned in Section III-E, we use a triangular accelera-
the intersection. Additionally, the outliers, which have been tion probability distribution as an example for modeling a more
already identified as not detected collisions, are visible as high realistic (but certainly not the real) driver behavior.
probability dots close to the intersection and diagonal. Fig. 7(c) Fig. 8 shows again the maximum reached collision proba-
shows very low estimated collision probabilities for N O C RASH bility per approaching vehicle. When comparing these distri-
approaches. butions in different situations for the beacon interval of 1 s
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 at 07:06:20 UTC from IEEE Xplore. Restrictions apply.
 JOERER et al.: VEHICULAR NETWORKING PERSPECTIVE ON ESTIMATING VEHICLE COLLISION PROBABILITY AT INTERSECTIONS 1809

Fig. 9. Worst case collision probabilities per bin calculated based on beacons received using a beaconing interval of 0.04 s; the dot size and color represent
the collision probability of the bin by showing large dots for high collision probabilities. (a) Minimum, uniform distribution, C RASH . (b) Minimum, triangular
distribution, C RASH . (c) Maximum, triangular distribution, N O C RASH .

with Fig. 6(d), we notice that the distribution of the triangular

distribution is more compact for the outcome C RASH, whereas
the other two outcomes have similar distributions. Therefore,
with a more realistic behavior (triangular distribution), we see
fewer false negatives.
Another positive aspect in a more realistic setup, i.e., using
the triangular distribution, is shown in Fig. 9. Here, again,
Fig. 10. Evolution of the collision probability for a typical C RASH intersec-
the correlation between sender/receiver distances and collision tion approach.
probabilities is depicted. When comparing Fig. 9(a) and (b),
it turns out that the collision probability estimation allows the if the channel becomes overloaded. Both effects are considered
prediction of a future crash already at larger distances from the in our simulation study as well. Moreover, the step height of the
intersection. This would allow earlier intervention in critical collision probability strongly depends on the interval at which
situations. (We have already shown that there is no negative new information is being received.
effect for the N O C RASH outcome either.) Assuming that an approach has been identified as unavoid-
We plotted the minimum reached collision probability per able crash, i.e., the collision probability reached 100%, all
bin to demonstrate how the collision probability is behaving in future beacons will also yield a collision probability of 100%.
the worst case, although the same effects are also visible when Therefore, any action to prevent a crash needs to be triggered at
plotting the mean (data not shown). Plotting this minimum least one beacon prior to receiving one yielding 100%. We call
collision probability reveals some low collision probabilities this the Last Before Unavoidable (LBU) beacon.
close to the borders [in Fig. 9(a) and (b)]. We verified that the In the following, we investigate the collision probability
outliers are caused by intersection layout inaccuracies. estimation that has been calculated for LBU beacons. Fig. 11
To analyze the worst case results for outcome N O C RASH, shows the empirical complementary cumulative distribution
the maximum reached collision probability per bin for the function (eCCDF) of the calculated collision probability for
triangular distribution is plotted in Fig. 9(c). Although the LBU beacons, depending on the configured beacon interval.
collision probability values are higher compared with the mean Looking at Fig. 11(a), which shows the eCCDF for uniform
[cf. Fig. 7(c)], the plot confirms the trend that, close to the distribution of possible trajectories, it is shown that the median
intersection, the collision probability is decreasing and showing LBU collision probability estimation was below 75% and 50%
very low probabilities of around 5%. for slow beacon intervals of 0.5 and 1 s, respectively. This
would already necessitate a fairly small threshold for automated
reactions. Even worse, considering safety applications, it is
D. Implications on Vehicular Networking Concepts
necessary to cover almost all possible situations; thus, the 95th
After showing the applicability and validity of our collision and 99th success percentiles (also listed in Table III) are a better
probability estimation, we are ready to draw first conclusions indication of how a reaction threshold would need to be chosen.
related to vehicular networking concepts. Here, it becomes clear that slow beacon intervals (1 and 0.5 s)
Fig. 10 shows a typical evolution of the collision probability are not suitable because the implied reaction thresholds for
over time for a C RASH approach, i.e., for one vehicle approach- these beacon intervals are as low as 21% and 48%, thus leading
ing an intersection and finally colliding with another vehicle. to many false positives. Depending on the driver behavior
Two effects are immediately apparent. The collision probability (as illustrated by the triangular acceleration probability distri-
can only be updated when a beacon has been received; and it bution chosen for Fig. 11(b) and Table III), the necessary reac-
remains unchanged in the case of lost beacons. Loss of beacons tion threshold might be chosen some percentage points higher
will, for example, occur if radio communication is obstructed or (from 39% to 76%), but it still remains low.
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 at 07:06:20 UTC from IEEE Xplore. Restrictions apply.
 1810 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 4, MAY 2014

Fig. 11. eCCDF of collision probability for LBU beacons and approaches in group C RASH. (a) Uniform acceleration probability distribution. (b) Triangular
acceleration probability distribution.

TABLE III ing is not as effective as anticipated in suburban environ-

R EACTION T HRESHOLDS BASED ON LBU C OLLISION P ROBABILITY
ments. The presented LBU collision probability can be used
to further examine the behavior of information dissemination
protocols for intersection safety applications. Moreover, the
analysis of the maximum change of the collision probability
between two consecutive beacons can give new insights into
how to properly design information dissemination for such
applications.
Finally, the presented general approach of the intersection
collision probability can be extended to capture more details
For smaller beacon intervals (0.04 and 0.1 s), we see that such as lateral movements of vehicles or longitudinal vehicle
the necessary reaction thresholds are appreciably high (up to dynamics. Additional future work includes the investigation
99.3% assuming a triangular acceleration probability distribu- of more realistic (maybe even dependent) probability distri-
tion, when targeting a 95th percentile success rate and using bution functions for trajectories of two approaching vehicles.
the fastest beaconing interval). However, these values are only These probability distributions might also make use of machine
applicable for the highly idealistic network conditions assumed learning and hence might be able to predict possible future
in the simulation. In earlier work [3], we were able to show that trajectories more accurately on a per driver basis.
fixed rate beaconing with such high rates would overload the
network, leading to excessive packet loss, which, as discussed
earlier, would have fatal implications for ICWSs.
R EFERENCES
[1] G. Karagiannis, O. Altintas, E. Ekici, G. Heijenk, B. Jarupan, K. Lin, and
VI. C ONCLUSION T. Weil, “Vehicular networking: A survey and tutorial on requirements, ar-
chitectures, challenges, standards and solutions,” IEEE Commun. Surveys
In conclusion, we were able to define a collision probability Tuts., vol. 13, no. 4, pp. 584–616, Jul. 2011.
[2] F. Dressler, F. Kargl, J. Ott, O. K. Tonguz, and L. Wischhof, “Re-
estimation scheme that allows assessment of the criticality of search challenges in inter-vehicular communication—Lessons of the 2010
an intersection approach based on exchanged beacons, such Dagstuhl seminar,” IEEE Commun. Mag., vol. 49, no. 5, pp. 158–164,
as CAMs or BSMs. Given information about two approaching May 2011.
[3] C. Sommer, O. K. Tonguz, and F. Dressler, “Traffic information systems:
vehicles (such as their current position and speed), we are Efficient message dissemination via adaptive beaconing,” IEEE Commun.
able to derive potential future trajectories and calculate the Mag., vol. 49, no. 5, pp. 173–179, May 2011.
probability of a crash. In general, this value can be subsequently [4] Intelligent Transport Systems (ITS); Vehicular Communications; Basic Set
of Applications; Part 2: Specification of Cooperative Awareness Basic
exploited for warnings to the human driver, although this is Service, ETSI, TS 102 637-2 V1.1.1, Apr. 2010.
not the focus of this paper. Instead, we evaluated the impact [5] C. Sommer, S. Joerer, M. Segata, O. K. Tonguz, R. Lo Cigno, and
of beaconing-based IVC on whether the calculated collision F. Dressler, “How shadowing hurts vehicular communications and how
dynamic beaconing can help,” in Proc. 32nd IEEE INFOCOM, Turin,
probability can be used for assessing the criticality of situations. Italy, Apr. 2013, pp. 110–114.
In an extensive number of simulation experiments, we validated [6] S. Biswas and F. Dion, “Vehicle-to-vehicle wireless communication proto-
the proposed criticality metric and carefully evaluated the col- cols for enhancing highway traffic safety,” IEEE Commun. Mag., vol. 44,
no. 1, pp. 74–82, Jan. 2006.
lision probability estimation. [7] S. Joerer, M. Segata, B. Bloessl, R. Lo Cigno, C. Sommer, and F. Dressler,
Our findings show that our metric clearly reveals the critical- “To crash or not to crash: Estimating its likelihood and potentials of
ity of an intersection approach. We also show, using periodic beacon-based IVC systems,” in Proc. 4th IEEE VNC, Seoul, Korea,
Nov. 2012, pp. 25–32.
beaconing as an example, how the metric can be applied for [8] Wireless Access in Vehicular Environments, IEEE Std. 802.11p-2010,
evaluating IVC solutions and how simple periodic beacon- Jul. 2010.
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 at 07:06:20 UTC from IEEE Xplore. Restrictions apply.
 JOERER et al.: VEHICULAR NETWORKING PERSPECTIVE ON ESTIMATING VEHICLE COLLISION PROBABILITY AT INTERSECTIONS 1811

[9] M. Segata and R. Lo Cigno, “Emergency braking: A study of network [33] R. Verma and D. Del Vecchio, “Safety in semi-autonomous multi-vehicle
and application performance,” in Proc. 8th ACM Int. Workshop VANET, systems: A hybrid control approach,” IEEE Robot. Autom. Mag., vol. 18,
Las Vegas, NV, USA, Sep. 2011, pp. 1–10. no. 3, pp. 44–54, Sep. 2011.
[10] M. Ferreira, R. Fernandes, H. Conceição, W. Viriyasitavat, and [34] M. Hafner, D. Cunningham, L. Caminiti, and D. Del Vecchio, “Coop-
O. K. Tonguz, “Self-organized traffic control,” in Proc. 7th ACM Int. erative collision avoidance at intersections: Algorithms and experiments,”
Workshop VANET, Chicago, IL, USA, Sep. 2010, pp. 85–90. IEEE Trans. Intell. Transp. Syst., vol. 14, no. 3, pp. 1162–1175, Sep. 2013.
[11] T. Neudecker, N. An, O. K. Tonguz, T. Gaugel, and J. Mittag, “Feasi- [35] M. Hafner, D. Cunningham, L. Caminiti, and D. Del Vecchio, “Automated
bility of virtual traffic lights in non-line-of-sight environments,” in Proc. vehicle-to-vehicle collision avoidance at intersections,” presented at the
9th ACM Int. Workshop VANET, Low Wood Bay, U.K., Jun. 2012, 18th ITS World Congress, Orlando, FL, USA, Oct. 2011, Paper 1314.
pp. 103–106. [36] C. Sommer, R. German, and F. Dressler, “Bidirectionally coupled network
[12] S.-H. Chang, C.-Y. Lin, C.-C. Hsu, C.-P. Fung, and J.-R. Hwang, “The and road traffic simulation for improved IVC analysis,” IEEE Trans.
effect of a collision warning system on the driving performance of young Mobile Comput., vol. 10, no. 1, pp. 3–15, Jan. 2011.
drivers at intersections,” Transp. Res. F, Traffic Psychol. Behav., vol. 12, [37] M. Treiber, A. Hennecke, and D. Helbing, “Congested traffic states in
no. 5, pp. 371–380, Sep. 2009. empirical observations and microscopic simulations,” Phys. Rev. E, Stat.
[13] H. Berndt, S. Wender, and K. Dietmayer, “Driver braking behavior during Nonlin. Soft Matter Phys., vol. 62, no. 2, pp. 1805–1824, Aug. 2000.
intersection approaches and implications for warning strategies for driver [38] D. Schrauben and J. Flegel, “Police vehicle evaluation model year 2013,”
assistant systems,” in Proc. IEEE IV Symp., Istanbul, Turkey, Jun. 2007, Michigan State Police, Precision Driving Unit, Lansing, MI, USA, Tech.
pp. 245–251. Rep., Dec. 2012.
[14] T. Mangel, F. Schweizer, T. Kosch, and H. Hartenstein, “Vehicu- [39] C. Sommer, D. Eckhoff, R. German, and F. Dressler, “A computationally
lar safety communication at intersections: Buildings, non-line-of-sight inexpensive empirical model of IEEE 802.11p radio shadowing in urban
and representative scenarios,” in Proc. 8th Int. Conf. WONS Services, environments,” in Proc. 8th Int. Conf. WONS Services, Bardonecchia,
Bardonecchia, Italy, Jan. 2011, pp. 35–41. Italy, Jan. 2011, pp. 84–90.
[15] H. Chen, L. Cao, and D. B. Logan, “Investigation into the effect of an [40] D. Eckhoff, C. Sommer, and F. Dressler, “On the necessity of accurate
intersection crash warning system on driving performance in a simulator,” IEEE 802.11p models for IVC protocol simulation,” in Proc. 75th IEEE
Traffic Inj. Prev., vol. 12, no. 5, pp. 529–537, Oct. 2011. VTC-Spring, Yokohama, Japan, May 2012, pp. 1–5.
[16] D. Valerio, F. Ricciato, P. Belanovic, and T. Zemen, “UMTS on the road:
Broadcasting intelligent road safety information via MBMS,” in Proc.
67th IEEE VTC-Spring, May 2008, pp. 3026–3030.
[17] C. Sommer, A. Schmidt, Y. Chen, R. German, W. Koch, and F. Dressler,
“On the feasibility of UMTS-based traffic information systems,” Ad Hoc
Netw., vol. 8, no. 5, pp. 506–517, Jul. 2010.
[18] R. K. Schmidt, T. Leinmüller, E. Schoch, F. Kargl, and G. Schäfer, “Ex- Stefan Joerer (S’12) received the B.Sc. and M.Sc.
ploration of adaptive beaconing for efficient intervehicle safety communi- degrees in computer science from the University of
cation,” IEEE Netw. Mag., vol. 24, no. 1, pp. 14–19, Jan./Feb. 2010. Innsbruck, Innsbruck, Austria, in 2009 and 2011,
[19] D. Rossi, R. Fracchia, and M. Meo, “VANETs: Why use beaconing at respectively. He is currently working toward the
all?” in Proc. IEEE ICC, May 2008, pp. 2745–2751. Ph.D. degree with the Computer and Communication
[20] DSRC Message Communication Minimum Performance Requirements: Systems Group, Institute of Computer Science, Uni-
Basic Safety Message for Vehicle Safety Applications, SAE, Draft Std. versity of Innsbruck.
J2945.1 Revision 2.2, Apr. 2011, SAE Int. DSRC Committee. His research interests include protocols and appli-
[21] D. Eckhoff, C. Sommer, R. German, and F. Dressler, “Cooperative aware- cations of inter-vehicular communications.
ness at low vehicle densities: How parked cars can help see through
buildings,” in Proc. IEEE GLOBECOM, Houston, TX, USA, Dec. 2011,
pp. 1–6.
[22] N. An, M. Maile, D. Jiang, J. Mittag, and H. Hartenstein, “Balancing the
requirements for a zero false positive/negative forward collision warn-
ings,” in Proc. 10th Int. Conf. WONS Services, Banff, Canada, Mar. 2013,
pp. 191–195.
[23] O. K. Tonguz, N. Wisitpongphan, and F. Bai, “DV-CAST: A distributed Michele Segata (S’11) received the B.Sc. and M.Sc.
vehicular broadcast protocol for vehicular ad hoc networks,” IEEE Wire- degrees in computer science from the University of
less Commun., vol. 17, no. 2, pp. 47–57, Apr. 2010. Trento, Trento, Italy, in 2009 and 2011, respectively.
[24] M. Werner, R. Lupoaie, S. Subramanian, and J. Jose, “MAC layer He is currently working toward the joint Ph.D. de-
performance of ITS G5—Optimized DCC and advanced transmitter gree with the Institute of Computer Science, Univer-
coordination,” in 4th ETSI TC ITS Workshop, Doha, Qatar, Feb. 2012, sity of Innsbruck, Innsbruck, Austria, and with the
pp. 1–21. Department of Information Engineering and Com-
[25] Intelligent Transport Systems (ITS); Decentralized Congestion Control puter Science, University of Trento, under the super-
Mechanisms for Intelligent Transport Systems Operating in the 5 GHz vision of Prof. F. Dressler and R. Lo Cigno.
Range; Access Layer Part, ETSI, Std. TS 102 687 V1.1.1, Jul. 2011. His current research interests include simulation
[26] L. Le, A. Festag, R. Baldessari, and W. Zhang, “Vehicular wireless short- models, network protocols, and management strate-
range communication for improving intersection safety,” IEEE Commun. gies for autonomous driving, with particular focus on platooning.
Mag., vol. 47, no. 11, pp. 104–110, Nov. 2009. Mr. Segata received a Ph.D. grant from the BIT School, which is a joint effort
[27] J. J. Haas and Y.-C. Hu, “Communication requirements for crash avoid- of the Bolzano, Innsbruck, and Trento local governments.
ance,” in Proc. 7th ACM Int. Workshop VANET, Chicago, IL, USA,
Sep. 2010, pp. 1–10.
[28] A. Tang and A. Yip, “Collision avoidance timing analysis of DSRC-based
vehicles,” Accid. Anal. Prev., vol. 42, no. 1, pp. 182–195, Jan. 2010.
[29] M. Brannstrom, E. Coelingh, and J. Sjoberg, “Model-based threat assess-
ment for avoiding arbitrary vehicle collisions,” IEEE Trans. Intell. Transp.
Syst., vol. 11, no. 3, pp. 658–669, Sep. 2010. Bastian Bloessl (S’12) received the Diploma in
[30] S. Lefèvre, C. Laugier, and J. Ibañez-Guzmán, “Risk assessment at road computer science from the University of Würzburg,
intersections: Comparing intention and expectation,” in Proc. IEEE IV Würzburg, Germany, in 2011.
Symp., Alcalá de Henares, Spain, Jun. 2012, pp. 165–171. Since 2011, he has been with the Computer and
[31] M. Liebner, M. Baumann, F. Klanner, and C. Stiller, “Driver intent infer- Communication Systems Group, Institute of Com-
ence at urban intersections using the intelligent driver model,” in Proc. puter Science, University of Innsbruck, Innsbruck,
IEEE IV Symp., Alcalá de Henares, Spain, Jun. 2012, pp. 1162–1167. Austria. He is currently working on software-defined
[32] H.-S. Tan and J. Huang, “DGPS-based vehicle-to-vehicle cooperative radio solutions for vehicular and wireless sensor
collision warning: Engineering feasibility viewpoints,” IEEE Trans. Intell. networks.
Transp. Syst., vol. 7, no. 4, pp. 415–428, Dec. 2006.
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 at 07:06:20 UTC from IEEE Xplore. Restrictions apply.
 1812 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 4, MAY 2014

Renato Lo Cigno (SM’11) received the Electronic Falko Dressler (SM’08) received the M.Sc. and
Engineering degree with specialization in telecom- Ph.D. degrees from the University of Erlangen,
munications from Politecnico di Torino, Torino, Erlangen, Germany, in 1998 and 2003, respectively.
Italy, in 1988. He was an Assistant Professor at the Com-
He then worked with Politecnico di Torino until puter Networks and Communication Systems Chair,
2002. In 1998 and 1999, he was a Visiting Scholar Department of Computer Science, University of
with the Department of Computer Science, Univer- Erlangen, where he was also coordinating the Au-
sity of California, Los Angeles, CA, USA. He is tonomic Networking Group. He is currently a Full
currently an Associate Professor with the Depart- Professor of computer science and the Head of
ment of Computer Science and Telecommunications, the Computer and Communication Systems Group,
University of Trento, Trento, Italy, where he leads a Institute of Computer Science, University of Inns-
research group in computer and communication networks. He is the author or bruck, Innsbruck, Austria. He is the author of the book Self-Organization in
co-author of more than 130 papers in international and peer-reviewed journals Sensor and Actor Networks (Chichester, U.K.: Wiley, 2007). He is actively par-
and conferences. His current research interests include performance evaluation ticipating in the Internet Engineering Task Force standardization. His research
of wired and wireless networks, modeling and simulation techniques, conges- interests include adaptive wireless networking and self-organization methods
tion control, peer-to-peer networks, and networked systems in general, with with applications in wireless ad hoc and sensor networks, inter-vehicular
specific attention toward applications and sustainable solutions. communications, bioinspired nanonetworking, and network security.
Mr. Lo Cigno is a member of the Association for Computing Machinery Dr. Dressler is a Senior Member of the Association for Computing Machin-
(ACM). He has served as General Chair for the IEEE International Conference ery (ACM). He has served as Co-Chair for the Technical Program Committees
on Peer-to-Peer Computing and General Chair and Technical Program Com- (TPCs) of the IEEE Vehicular Networking Conference, the IEEE Vehicular
mittee (TPC) Chair for the ACM International Workshop on Wireless Mobile Technology Conference, and the IEEE Global Communications Conference;
Applications and Services on WLAN Hotspots in different years. He has also the Area TPC Chair for the IEEE Conference on Computer Communications;
served on many TPCs of IEEE and ACM conferences. He has served as an Area and the Poster/Demo Chair for the ACM Annual International Conference
Editor for Computer Networks. on Mobile Commmunications. He is an IEEE Distinguished Lecturer in the
fields of inter-vehicular communications, self-organization, and bioinspired
nanonetworking. He has served as a Guest Editor for special issues on self-
Christoph Sommer (M’11) received the M.Sc. de- organization, autonomic networking, and bioinspired communications for the
gree in computer science and the Ph.D. degree (with IEEE J OURNAL ON S ELECTED A REAS IN C OMMUNICATIONS, Elsevier Ad
distinction) in engineering from the University of Hoc Networks, and others. He serves as an Editor for journals such as the
Erlangen, Erlangen, Germany, in 2006 and 2011, IEEE T RANSACTIONS ON M OBILE C OMPUTING, Elsevier Ad Hoc Net-
respectively. works, ACM/Springer Wireless Networks, and Elsevier Nano Communication
In 2010, he was a Visiting Scholar with the Networks.
Research Group of O. Tonguz, Department of
Electrical and Computer Engineering, Carnegie
Mellon University, Pittsburgh, PA, USA. In 2012,
he was a Visiting Scholar with the Research Group
of M. Gerla, Department of Computer Science, Uni-
versity of California, Los Angeles, CA, USA. He is currently a Postdoctoral
Researcher with the Computer and Communication Systems Group, University
of Innsbruck, Innsbruck, Austria. His research interests include traffic efficiency
and safety and security aspects of car-to-x communications in heterogeneous
environments.
Dr. Sommer has been a member of the ACM/Springer Wireless Networks
Editorial Board since 2011.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on May 03,2024 at 07:06:20 UTC from IEEE Xplore. Restrictions apply.