Communication-base Collision Avoiance between Vulnerable Roa Users an Cars Michele Segata, Romas Vijeikis, Renato Lo Cigno Dept. of Information Engineering an Computer Science, University of Trento, Italy romas.vijeikis@stuenti.unitn.it, {msegata,locigno}@isi.unitn.it Abstract The avent of wireless communication for vehicles paves the roa for a bounty of cooperative applications: The most interesting being cooperative safety awareness. By exchanging information vehicles can become aware of each other an prevent angerous situations that can lea to crashes either by early warning rivers or by automatic vehicle control, a solution particularly appealing for self-riving cars. While research on vehicular safety mainly focuses on vehicle-to-vehicle safety, we can exploit communication to implement applications aime at protecting vulnerable roa users, such as peestrians or cyclists. In this paper, we start by exploiting a probability framework for estimating the likelihoo of collision between a vehicles approaching an intersection an a cyclist, in light of the feasibility of communication between the two. On top of this framework, we esign a simple application that, uner certain conitions, informs the car river (assume to have to yiel preceence) of the possible collision. We moel the reaction of the river to the warning an analyze possible benefits an rawbacks of such an application. The contribution is not the application itself, which is obvious, but the insights in the results in light of communication capabilities an human reaction that provie a specific set of aspects that shoul be consiere in the esign of such a safety system. I. INTRODUCTION Vehicular communication is fostering a wie range of research ieas that span from safety to efficiency, as well as comfort-oriente applications. Starting from the last category, we can list examples such as automatic toll collection, service announcements, an multimeia content ownloa [1]; in practice, all the services that can improve the experience of rivers an passengers without a special focus to the vehicle itself. Efficiency applications, instea, focus on reucing travel times an fuel consumption, saving money an reucing carbon footprint of vehicles. Examples of these applications inclue optimal spee avisory systems for traffic lights [2], virtual traffic lights [3], an platooning [4] [6]. Safety, however, is probably the main river towar the evelopment of cooperative vehicular applications, as there is a huge accient an eath reuction potential. The list of propose vehicular safety applications is very large an inclues traffic signal violation, cooperative forwar collision warning, an emergency electronic brake lights to name a few [7], [8]. An important application, especially in urban environments, is intersection collision warning/assistance [9], [10]. The aim of the application is Romas Vijeikis work was supporte by Movalia (https://www.movalia.it/) an EIT Digital (https://www.eitigital.eu/). to mitigate the effects or, if possible, completely avoi an accient at an intersection ue to human mistakes or traffic violations. Its importance is funamental, as the majority of urban accients happen near or at an intersection. Vehicular applications initially focuse on car-to-car safety only, but there is no oubt that vulnerable roa users such as peestrians or cyclists are those who more often suffer fatal or serious injuries when involve in an accient: Car passengers are protecte by the vehicle itself (chassis, safety belt, airbags), while peestrian an cyclists are simply expose. To have an iea of the problem, out of 5.9 million crashes in the European Union in 2013, 2.6 million involve peestrians an cyclists [11]. For these reasons, in this paper we focus on an intersection collision avoiance application for cyclists. Inspire by the work in [12], which focuses on car-to-car collision avoiance, we analyze an intersection collision avoiance system aime at protecting cyclists from vehicles that may collie with them. We aapt the concept of collision probability in [12] to moel T-shape intersections, that are very common in areas where a bike lane on a major urban roa intersects a resiential roa. We moel misbehaving car rivers that, while approaching the intersection, ignore a cyclist right-of-way ue to, for example, a large object obstructing the line of sight. In this scenario, we analyze the collision rate without an with a warning application that exploits Bike-to-vehicle (B2V) communication to warn the river about a possible collision. We moel rivers reaction time to the warning, investigating the benefits an rawbacks of such an application uner ifferent parameters, obtaining interesting an useful insights for a realworl implementation. The main contribution of this paper is the analysis of the opportunity an timing of warning of the application in a scenario that takes into account most of the real constraints of a real scenario. II. BACKGROUND AND RELATED WORK Besies passive safety measures such as helmets, lightreflective clothes, an front an back lights that tries to avoi accients or reuce their effects, we also fin active safety systems installe in commercial cars. Some examples inclue autonomous emergency braking, which exploits a raar, a liar, or a camera to etect objects in collision course an autonomously brake the car [13].
Figure 1: T intersection layout with blin zone problem. While these systems can alreay reuce the risk of a collision or its consequences, they have an intrinsic limitation, which is their fiel of view. Sensor-base solutions become ineffective when there is no line of sight between the vehicle an the object in the collision course. As an example, take the situation epicte in Figure 1. The picture represents a T intersection where the main roa is the one going top to bottom an then turning to the right, inee a left turn for a bike following the blue line trajectory. Vehicles coming from the left shoul give the right-of-way to bikes on the bike-lane of the main roa. This layout is angerous because the river of a vehicle coming from the left might think of being on the main roa just because it is straight, or simply being istracte by whatever reason. Such a mistake can be mitigate by sensor-base safety applications, but not when an object obstructs the line of sight: in this case sensors are no better than human eyes. The alternative to pure sensor-base solutions is cooperation by means of communication. By actively exchanging ata, vehicles an other roa actors can get informe about each others presence, preventively compute the risk of a collision, an, if neee, take countermeasures. Classical Vehicle-to-vehicle (V2V) applications consier the use of an IEEE 802.11-base technology (i.e., IEEE 802.11p [14]) as safety mainly requires local communication. In aition, other technologies such as Visible Light Communication (VLC) [15] or 4G cellular (Long Term Evolution (LTE)) [16] have been propose for the purpose. In this paper we consier classical V2V using IEEE 802.11p. In the literature we fin some bicycle safety-relate projects that investigate the issue of vulnerable roa users. One is the WATCH-OVER project [17], which focuse on several aspects concerning the protection of vulnerable roa users. The project investigate ifferent scenarios, technologies, algorithms, an applications. Another project is VRUITS [18]. This project also ientifie an investigate critical scenarios for vulnerable roa users, showing that intersections are the most angerous accient locations, especially in poor visibility conitions. In aition, the project took into account the human factor, e.g., eveloping intelligent traffic signals that take into account the time require to cross a street or an intersection. BikeCOM [19], instea, focuses more on the evelopment of communication strategies for cyclists an vehicles base on smartphones. The aim of the evelope application was to establish a B2V connection between smartphones use in vehicles an by vulnerable roa users to exchange safety relate information. Finally, we have the PROSPECT project [20]. The project is still ongoing an focuses on the evelopment of active safety systems implementing inepenent emergency braking, focusing on better etection algorithms as well. In this paper, rather than using an experimental approach, we focus on moeling. In particular, we moel the misbehavior of vehicle rivers approaching an intersection as epicte in Figure 1. Position, spee an acceleration ata are share between cars an bikes using IEEE 802.11p, an the collision probability between a vehicle an a bicycle in a T-shape intersection is compute extening [12]. Base on this probability, we efine a threshol-base mechanism to inform the river of the car about the incoming bicycle in collision course, simulating its reaction to the warning. We then stuy the impact of choosing ifferent probability threshols an istances to the intersection for issuing the warning. III. PROTECTING VULNERABLE ROAD USERS A. Computing a Collision Probability We exten the formulation of collision probability presente in [12] to inclue a T-shape intersection. The whole formulation of the collision probability can be foun in [12]. For our purposes it is sufficient to know that the concept of collision probability consiers all possible trajectories that the the car an the bike can take while approaching the intersection exploiting the ata share via communication. The fraction of trajectories that le to a collision over all trajectories gives the probability of collision. The computation consiers the following parameters of two vehicles A an B approaching the intersection: spees (v A, v B ), current istance to the intersection of the paths ( A, B ), maximum acceleration an eceleration (a Amax, a Amin, a Bmax, a Bmin ) an current acceleration (a A, a B ). The istances A an B efine the istance travele by vehicles following their paths to the center point of intersection of their routes. These paths are not necessarily straight lines an epen on the topology on the roas leaing to the junction. Base on these parameters, we can reconstruct all possible feasible trajectories an we compute the probability of collision base on how many of those overlap. For the sake of completeness, we briefly list the formula for computing the probability of collision P c : P c = amax a min p(a B ) amax a min p(a A )coll( A v A a A, B v B a B )a A a B (1)
The formula, given the initial istance to the intersection an the spee of the vehicle, computes all possible trajectories obtaine by choosing an acceleration in the range [a min, a max ] an check whether they woul lea to a collision or not. In aition, the formula weights each single trajectory with the probability of choosing that specific acceleration. The computation can thus be customize by using ifferent probability moels. To test for collision we check whether the two vehicles are insie the potential collision area at the same time, an this is one by computing the time at which a vehicle enters an leaves the collision area (t e an t l, respectively). The enter an leave times can be compute by consiering the istances on the trajectory that correspon to the points where the vehicle enters an leaves the collision area (see Figure 2 for a graphical explanation). We can efine these istances as e an l, an, as originally one in [12] by solving the following equation e/l = a 2 t2 e/l + vt e/l (2) we can compute t e an t l as { v2 +2 a e/l v + t e/l = a, if v 2 + 2 a e/l 0, otherwise Equation (3), however, consiers also cruising spees which, especially for a bicycle, woul be impossible. Differently to [12] we introuce a limit on the maximum spee reachable by vehicles an bikes. In particular we assume maximum spees v a = 72 km/h an v b = 50 km/h for cars an bicycles, respectively. To account for maximum spees we nee to change Equation (3). First, we compute the time at which the vehicles reaches the maximum spee, i.e., { vmax v t vmax = a, if a > 0 (4) 0, otherwise Then we compute the istance covere by the vehicle while reaching the maximum spee an the remaining istance to travel as vmax = v a t2 v max, after = e/l vmax. (5) 2 Finally, we can compute t e an t l accounting for maximum spee as t e/l = { t e/l (Equation (3)), t vmax + after v max, if t e/l t vmax otherwise. The final point is efining the istribution for the probabilities p(a A ) an p(a B ) in Equation (1). In this work we raw the probability of acceleration from a triangular istribution having the minimum an maximum values set to a min an a max, an the moe set to the current acceleration. B. Moeling Car Drivers Behavior We consier the river of the car to be the only actor in the scenario that can prevent the accient. The biker continues straight through the intersection because of the right of way (3) (6) Aenter Benter Aleave Bleave Figure 2: An example of ientification of e an l positions for both vehicles on the intersection. an it is the river of the car, who receives a waning from the application an has to brake to avoi it. We perform the analysis using the Veins vehicular networking framework [21], which couples the OMNeT++ network simulator with the SUMO vehicular mobility simulator. SUMO inclues behavioral moels that implement traffic rules, such as right of way, but it oes not moel mistakes an it is completely accient-free. To our purposes we nee to moify the behavioral moel, in this case the Intelligent Driver Moel (IDM), to ignore traffic rules. This is one by isabling right of way rules when the vehicle is within a few meters to the intersection. In aition, we moel the reaction of the river to warnings issue by the system. The bike perioically sens packets incluing its own position an spee. At each receive upate, the vehicle computes the collision probability P c an informs the river if P c is above a certain threshol an if the vehicle is within a certain istance to the intersection. The rationale of the istance threshol is to limit the number of false positives, which can give the river the impression that the application is not working. Both probability an istance threshols are object of stuy in this paper. When a warning is notifie to the river, the latter reacts by ecelerating with maximum eceleration a min own to a stop. We also consier a reaction time of 1 s, so the river oes not immeiately starts braking when receiving a warning. IV. PERFORMANCE EVALUATION As anticipate in Section III-B we use the Veins vehicular networking simulation framework to analyze the performance of the propose collision avoiance system in terms of safety. The simulation coe is available upon request from the authors. First, we run a set of simulations where car rivers ignore the right of way an our collision avoiance application is isable. This enables us to obtain collisions between vehicles an bicycles. In a secon set of runs we enable our safety application measuring again the number of collisions. In the simulation we inject pairs of cars an bicycles with ranomize characteristics (Table I) to obtain a wie range of
Density (%) 100 80 60 40 20 0 0.0 0.2 0.4 0.6 0.8 1.0 P c Safe Pass Crash Near Miss Figure 3: Distribution of compute collision probabilities 5 m from the intersection, safety system isable. situations, which inclue safe passes, near misses, an actual crashes. Near misses are a special kin of safe passes where the bike an the car get very close to each other. In this context, with very close we mean 2 m. With respect to communication, we consier an IEEE 802.11p network. Although it is not yet clear which kin of communication technology will support safety applications for vulnerable roa users, we can assume this will be a local network available on user smartphones, such as Wi-Fi, LTE Direct, or 802.11p. In the context of the paper, however, the technology is irrelevant as our contribution is more application-oriente, an consiering realistic communication as well as the effects of impairments is out of our scope. For the sake of completeness, Table II lists communication parameters use in the simulation. Finally, we stuy the behavior of the system uner ifferent triggering conitions, in particular by choosing ifferent collision probability threshols an ifferent minimum istances to the intersection. With respect to the probability threshol we consier the full range, i.e., from 0 to 1, while for istances to the intersection we consier the range between 21 m an 14 m, in steps of 1 m. In total, we simulate 850 approaches per each set of parameters. We choose the parameters to intentionally esign angerous intersection approaches that can lea to a collision, a conition that is necessary for the evaluation of the system. Collision rates thus are not comparable to real-worl situations. Table I: IDM parameters for bicycles an cars use in the evaluation. Parameter Cars Bicycles a min (m/s 2 ) U( 6.5, 8.8) U( 2.8, 3.5) a max (m/s 2 ) U(2.8, 3.5) U(1.2, 1.5) v max (m/s) U(8.5, 17) U(4.5, 5.5) v (m/s) U(0, v max) U(0, v max) A. Collision Probability Overview The first step of our evaluation is reasoning on the behavior of the collision probability, easing the performance analysis of the application. Figure 3 shows the histogram of compute P c values when the car is between 5 m an the center of the Density (%) 100 80 60 40 20 0 Table II: Communication parameters. Parameter Value Path loss moel Free space (α = 2.0) Frequency 5.89 GHz Banwith 10 MHz Bit Rate 18 Mbit/s Transmit power 20 mw Noise floor 110 Bm Minimum sensitivity 89 Bm PHY moel IEEE 802.11p MAC moel 1609.4 single channel (CCH) Beacon rate 5 Hz Safe Pass Crash Near Miss 0.0 0.2 0.4 0.6 0.8 1.0 Pc Figure 4: Distribution of compute collision probabilities 20 m from the intersection, safety system isable. intersection. In this scenario the collision avoiance assistance system is isable. The histogram shows a clear separation between probabilities compute uring safe pass an crash scenarios. In particular, P c for safe passes is always smaller than 0.25, while crashes result in values of P c larger than 0.5, with some rare cases below 0.4. The istribution of values for near misses is more wiesprea, ranging from 0.1 to 1.0. In the majority of the cases, however, near misses result in values of P c larger than 0.9. This first observation shows that P c correctly classifies intersection approaches, but it oes not give a feeling about how effective it can be in reucing intersection collisions. In this first example P c is compute 5 m to the intersection, where a collision might alreay be unavoiable. By taking the same measure 20 m from the intersection, however, the results change substantially (Figure 4). In particular, the range of P c values shrinks, becoming no larger than 0.7. In aition, values of P c for safe passes an crashes overlap, inicating that using P c as input for a collision avoiance system is not a trivial task, as we can easily prouce false positives an false negatives. Figure 5 explains the large ifference between Figures 3 an 4 by plotting the evolution of P c as function of the istance to the intersection for three intersection approaches, two of which le to a safe pass while the thir to a crash. The plot shows that, 30 m from the intersection, istinguishing between the three situations is harly possible, because of the large range of possible ecisions that can still be mae by the river. For one safe pass (green ash-otte line), P c raises until 30 m to the intersection, where it finally rops to 0 20 m to the intersection, correctly ientifying the absence of risk. For the
Pc 1.0 0.8 0.6 0.4 0.2 0.0 SafePass Safe pass with steep progression of collision estimation Typical crash 100 80 60 40 20 0 (m) Warning rates 0.8 0.6 0.4 0.2 0.0 false positive false negative 0.0 0.1 0.2 0.3 0.4 P c threshol Figure 5: Evolution of the collision probability for three ifferent intersection approaches. Collision rate 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 = 14 m = 15 m = 16 m = 17 m = 18 m = 19 m = 20 m = 21 m 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 P c threshol Figure 6: Collision rates as function of ifferent P c threshols an ifferent minimum istances to the intersection. secon safe pass (blue ashe line) P c is still comparable to the crash approach up until roughly 20 m to the intersection, where it starts ropping to 0. The crash approach correctly computes a P c value of 1, but only in the last 10 to 5 m. Figure 5 thus shows that the esign of such an application is non trivial, an can easily lea to misjugement, an thus false positives or false negatives. B. Intersection Collision Safety Evaluation The choice for investigating warning notifications istances between 21 m an 14 m is motivate by the results in Section IV-A. In this range, inee, the compute collision probability becomes meaningful an can be exploite by the safety system. In this section we thus stuy the reuction in collision rates, as well as the balance between false positives an false negatives. The reaer shoul bear in min that the results in this section o not represent the groun truth of what is achievable by an intersection collision avoiance system an that they epen on our specific implementation an moeling. The results represent foo for thoughts for the evelopment of this kin of systems. Figure 6 shows the collision rate as function of ifferent P c threshol an minimum istances to the intersection. For the sake of comparison, we also plot the collision rate for the approaches where the assistance application is isable, which, in this particular setup, is aroun 22 %. For large minimum istances (between 21 m an 18 m) an low P c threshols (up to 0.2), the application results in no crashes at all. As Figure 7: Comparison between false positives an false negatives as function of ifferent P c threshols, for a minimum warning istance of 20 m. we increase the P c threshol above 0.2, the application starts to loose performance approaching the no-assistance collision rate for a P c threshol that epens on the minimum warning istance. For shorter minimum warning istances, instea, the behavior changes. First we see that the collision rate for a P c threshol of 0 is not null, an the reason lies behin the fact that the vehicle might not have enough braking istance to stop or avoi the biker. In fact, as we increase the P c threshol, the collision rate ecreases before converging to the no-assistance collision rate. While this might seem non intuitive, the explanation is fairly simple. In some cases, by informing the river we actually cause a collision, which woul not happen otherwise because the car woul have crosse the intersection before the bike. This shows a funamental issue of these safety systems, that is when an if to inform a river at all. The graph suggests that the vehicle shoul never be in a situation where nothing can be one anymore, or where the situation can be mae worse. The system shoul not be responsible for accients, but it shoul only avoi or mitigate them. This suggests that warnings shoul not be triggere in such cases, but this statement is very ifficult to juge from a social an legal point of view. In aition, here we o not consier the impact force, which coul also be an interesting metric of evaluation. In any case this aspect must be accounte for in the esign of safety systems. The final question regaring the results in Figure 6 is the cost of having no crashes. For large minimum warning istances an low P c threshols we completely avoi crashes, but, as Figure 7 shows, the rate of false positives is high. The Figure also shows that instea false negatives can be reuce to zero maintaining the threshol P c large enough. To evaluate false negatives, we compute the amount of approaches that le to a crash or a near miss but where the warning was not issue to the river. The graph shows a high false positives rate, which is clearly an unesire feature in terms of marketing, as users experiencing large false positives rates might think the system oes not properly work an isable it. However, we have to keep in min that these results have been obtaine simulating only angerous situations. Inee, if we consier a rate of safe to angerous situation of 100 to 1 (very reasonable, otherwise
accients woul happen by the ay to every one of us) an a 20% of false positives, this means that there is one false positive every 500 times a car intersects with a bike on a T shape intersection, which is probably acceptable to users. Depening on esign goals, the P c threshol can be tune accoringly. We can choose to minimize the false positives rate while having no collisions at all, minimize the combination of false positives an false negatives, or minimize the false positives rate while keeping false negatives within an acceptable boun. For example, in this particular case setting a P c threshol of 0.3 maintains the false positives rate below 20 % an the collision rate (Figure 6) aroun 10 % halving the number of collisions. In conclusion, this evaluation highlights challenges an potentials of collision assistance systems showing that, uring the esign phase, we shoul also take into account social an legal aspects, an not only focusing on maximizing the safety gain. V. CONCLUDING DISCUSSION AND FUTURE WORK This work presente a preliminary stuy on the applicability of vehicular communication for protecting vulnerable roa users through an intersection collision avoiance application. We aapte the concept of collision probability in [12] to a T-shape intersection an exploite it to issue collision warning to a river. We moele the misbehavior of a car river an his/her reaction to a warning issue by the assistance system to iscover some funamental properties that shoul be accounte for in the esign phase. We showe that there are some funamental issues to be aresse. In particular, we nee to carefully choose the conitions for triggering a warning, in terms of collision probability an istance to the intersection. A low collision probability threshol can rastically reuce the collision rate, but at the cost of high false positives rate. In aition, the safety application shoul inform the riving within a reasonable istance to the intersection, so that there is still the possibility of avoiing a collision. The next steps inclue a more realistic moeling of both car rivers an cyclists. Simulation is an effective mean to stuy the impact of these applications on the safety of roa users, but the results must be built upon realistic an creible moels. This can be one for instance by consiering reconstructions of real worl accients (to moel misbehaving rivers) an by experimenting the human reaction to warnings while riving. Improving the moel with real ata we can isregar the trajectories that are improbable, improving the estimation of the collision likelihoo. On top of the moels an the best practices presente in this work we can start esigning effective applications aime at improving the safety of vulnerable roa users. REFERENCES [1] T. K. Mak, K. P. Laberteaux, an R. Sengupta, A Multi-channel VANET Proviing Concurrent Safety an Commercial Services, in 2n ACM International Workshop on Vehicular A Hoc Networks (VANET 2005). Cologne, Germany: ACM, September 2005, pp. 1 9. [2] D. Eckhoff, B. Halmos, an R. German, Potentials an Limitations of Green Light Optimal Spee Avisory Systems, in 5th IEEE Vehicular Networking Conference (VNC 2013). Boston, MA: IEEE, December 2013, pp. 103 110. [3] T. Neuecker, N. An, O. K. Tonguz, T. Gaugel, an J. Mittag, Feasibility of Virtual Traffic Lights in Non-Line-of-Sight Environments, in 9th ACM International Workshop on Vehicular Internetworking (VANET 2012). Low Woo Bay, Lake District, UK: ACM, June 2012, pp. 103 106. [4] C. Bergenhem, Q. Huang, A. Benmimoun, an T. Robinson, Challenges of Platooning on Public Motorways, in 17th Worl Congress on Intelligent Transport Systems (ITS 2010), Busan, Korea, Oct. 2010. [5] S. Santini, A. Salvi, A. S. Valente, A. Pescapè, M. Segata, an R. Lo Cigno, A Consensus-base Approach for Platooning with Inter- Vehicular Communications, in 34th IEEE Conference on Computer Communications (INFOCOM 2015). Hong Kong, China: IEEE, Apr. 2015, pp. 1158 1166. [6] M. Segata, Safe an Efficient Communication Protocols for Platooning Control, PhD Thesis (Dissertation), University of Innsbruck, February 2016. [7] National Highway Traffic Safety Aministration, Vehicle Safety Communications Project Task 3 Final Report: Ientify Intelligent Vehicle Safety Applications Enable by DSRC, NHTSA, Project Report DOT HS 809 859, March 2005. [8] M. Segata an R. Lo Cigno, Automatic Emergency Braking - Realistic Analysis of Car Dynamics an Network Performance, IEEE Transactions on Vehicular Technology, vol. 62, no. 9, pp. 4150 4161, Oct. 2013. [9] L. Le, A. Festag, R. Balessari, an W. Zhang, Vehicular Wireless Short-Range Communication for Improving Intersection Safety, IEEE Communications Magazine, vol. 47, no. 11, pp. 104 110, November 2009. [10] S. Joerer, B. Bloessl, M. Segata, C. Sommer, R. Lo Cigno, A. Jamalipour, an F. Dressler, Enabling Situation Awareness at Intersections for IVC Congestion Control Mechanisms, IEEE Transactions on Mobile Computing, vol. 15, no. 7, pp. 1674 1685, June 2016. [11] Roa Safety in the European Union: Trens, Statistics an Main Challenges, European Union, Report MI-AC-15-001-EN-N, March 2015. [12] S. Joerer, M. Segata, B. Bloessl, R. Lo Cigno, C. Sommer, an F. Dressler, A Vehicular Networking Perspective on Estimating Vehicle Collision Probability at Intersections, IEEE Transactions on Vehicular Technology, vol. 63, no. 4, pp. 1802 1812, May 2014. [13] D. Hyn, M. McCarthy, J. Carroll, M. Seil, M. Ewars, C. Visvikis, M. Tress, N. Ree, an A. Stevens, Benefit an Feasibility of a Range of new Technologies an Unregulate Measures in the Fiel of Vehicle Occupant Safety an Protection of Vulnerable Roa Users, European Union, Report NB-07-14-108-EN-N, March 2015. [14] IEEE, Wireless Access in Vehicular Environments, IEEE, St 802.11p- 2010, Jul. 2010. [15] M. Segata, R. Lo Cigno, H.-M. Tsai, an F. Dressler, On Platooning Control using IEEE 802.11p in Conjunction with Visible Light Communications, in 12th IEEE/IFIP Conference on Wireless On eman Network Systems an Services (WONS 2016). Cortina Ampezzo, Italy: IEEE, January 2016, pp. 124 127. [16] S. Kato, M. Hiltunen, K. Joshi, an R. Schlichting, Enabling Vehicular Safety Applications over LTE Networks, in 2n International Conference on Connecte Vehicles & Expo (ICCVE 2013), Las Vegas, NV, USA, December 2013, pp. 747 752. [17] L. Anreone an G. Wanielik, Vulnerable Roa Users Thoroughly Aresse in Accient Prevention: The WATCH-OVER European Project, in 14th Worl Congress on Intelligent Transport Systems (ITS 2007). Beijing, China: ITS Japan, October 2007, pp. 629 636. [18] J. Scholliers, D. Bell, A. Morris, an A. B. García, Improving Safety an Mobility of Vulnerable Roa Users through ITS Applications, in 5th Transport Research Arena (TRA 2014). Paris, France: IFSTTAR, April 2014. [19] P. Gustafsson, J. Muñez, L. Lingren, C. Boa, an M. Dozza, BikeCOM: Cooperative Safety Application Supporting Cyclists an Drivers at Intersections, in 3r International Conference on Driver Distraction an Inattention, Gothenburg, Sween, September 2013. [20] I. Gohl, A. Schneier, J. Stoll, M. Wisch, an V. Nitsch, Car-to-cyclist Accients from the Car Driver s Point of View, in 5th International Cycling Safety Conference (ICSC2016), Bologna, Italy, November 2016. [21] C. Sommer, R. German, an F. Dressler, Biirectionally Couple Network an Roa Traffic Simulation for Improve IVC Analysis, IEEE Transactions on Mobile Computing, vol. 10, no. 1, pp. 3 15, Jan. 2011.