COOPERATIVE POSITIONING TECHNIQUES AND ALGORITHMS FOR LAND MOBILE APPLICATIONS

Transcription

1 COOPERATIVE POSITIONING TECHNIQUES AND ALGORITHMS FOR LAND MOBILE APPLICATIONS Azmir Hasnur Rabiain 1, Allison Kealy 2, Guenther Retscher 3, Nima Alam 4, Andrew Dempster 5, Dorota Brzezinska 6, Charles Toth 7, Vassilis Gikas 8 1 PhD Candidate, 2 Associate Professor, Department of Infrastructure Engineering The University of Melbourne, Parkville, Victoria, Australia 1 azmirhr@unimelb.edu.au 2 a.kealy@unimelb.edu.au 3 Professor, Department of Geodesy and Geoinformation Vienna University of Technology, Karlsplatz, Vienna, Austria Guenther.retscher@tuwien.ac.at 4 Research Associate, 5 Professor, School of Surveying and Geospatial Engineering University of New South Wales, Kensington, New South Wales, Australia 4 nima.alam@gmail.com 5 a.dempster@unsw.edu.au 6 Professor, 7 Senior Research Scientist, Civil, Environmental and Geodetic Engineering Ohio State University, Columbus, Ohio, United States of America 6 grejner-brzezins.1@osu.edu 7 toth.2@osu.edu 8 Assistant Professor, School of Rural and Surveying Engineering National Technical University of Athens, Zographou, Athens, Greece vgikas@central.ntua.gr KEY WORDS: Cooperative Positioning, VANET, Assisted Localization, Integrated Systems ABSTRACT: The major industry growth sectors that rely on continuous positioning information to enhance efficiency and safety are broadly across transportation and location based services (LBS). Significantly, delivering a robust positioning capability in priority operates environments for these sectors (urban areas, indoors, etc) is a challenge for the Global Navigation Satellite Systems (GNSS) - the primary positioning technology for LBS and transportation. Cooperative positioning (CP) techniques leverage the availability of a communications infrastructure to share information and data between objects within a neighbourhood. Developed originally for use across wireless sensor networks, CP techniques offer a viable solution for improving positioning for land mobile applications. In this scenario, people and vehicles operate as sensors that form an ad hoc network, sharing information sensed locally but also using relative measurements between objects. This paper describes an initiative undertaken through the Working Group on Ubiquitous Positioning within the International Federation of Surveyors, Commission 5 and International Association of geodesy, Commission 4. It presents the architectures, scenarios and algorithms used traditionally for CP and identifies innovative concepts and directions for adapting and improving these algorithms using constraints and smart modelling techniques specific to the land mobile operates environment. 1. INTRODUCTION Maintaining the availability of a position solution is a significant challenge for many applications such as intelligent transportation systems (ITS) and location based services (LBS). One of the ways of obtaining positioning solutions is through the use of Global Navigation Satellite Systems (GNSS), which enables users to solve for their positions at a global scale with achievable accuracy between 5-15 m (Hofmann-Wellenhof and Lichtenegger, 2001). The major drawback of GNSS is that it requires an open sky view and often, it does not work in dense urban environments. Therefore, relying on GNSS alone may not always be sufficient for critical ITS applications such as vehicular collision avoidance system. Also, these applications typically need sub-metre accuracy and high update rates (> 10 Hz), which are not achievable using standalone GNSS solutions (Zador et al., 2000). Cooperative positioning (CP) is a technique which potentially can overcome the shortcomings of GNSS for ITS applications. CP is an approach where nodes or vehicles share information amongst themselves within a network in order to increase availability, integrity and continuity of positioning. The information shared can be in the form of inter-vehicle ranges, relative speed, orientation, and satellite related data. Sharing information will help the vehicles

2 within the network to obtain positioning solutions even when the requirement of GNSS positioning is not met. A network of vehicles sharing information is termed as Vehicular Ad-hoc Networks (VANET). One of the ways to establish communication in a VANET is through the use of Dedicated short-range communications (DSRC), which is a wireless communication channel designed specifically for vehicle-vehicle (V-V) and vehicle-infrastructure (V-I) communications. In the U.S., the Federal Communication Commission (FCC) has allocated DSRC with a dedicated bandwidth of 75 MHz in the GHz band, whereas the European Telecommunications Standards Institute (ETSI) has allocated a dedicated bandwidth of 30 MHz in the 5.9 GHz band. Some of the planned applications of DSRC includes intelligent transportation system (ITS), traffic management, safety and efficiency and as it is able to provide low latency, high speed communication, and strong and relative close proximity signals (Parker and Valaee, 2007), hence making it a suitable candidate for the enablement of CP in VANET. The work presented in this paper is an initiative of a joint working group on Ubiquitous Positioning within the International Federation of Surveyors (FIG), Commission 5 and International Association of Geodesy (IAG), Commission 4. Some of the focuses of the working group are performance characterization of positioning sensors technologies that can play a role in the development of ubiquitous positioning systems, theoretical and practical evaluation of current algorithms for measurement integration, development of new integration algorithms and innovative modelling techniques and generates formal parameters that describe the performance of emerging technologies. This paper will only discuss CP specific for VANET. Firstly, it will present the types of measurements and their viability to be used in realising CP using DSRC. Then, the subsequent section details the current and planned work undertaken by the joint FIG commission 5 and IAG commission 4 in realising and advancing CP for land mobile vehicles. Finally, it concludes the paper in section COLLABORATIVE POSITIONING TECHNIQUES CP in VANET has received great attention over recent years as relying on GNSS standalone system could not meet the requirements of ITS. CP can be broadly categorized in terms of sharing of ranges or range-rates and sharing satellite related information between V-V and V-I. The CP techniques presented in the following will only focus on range/range-rates based CP and their viability for VANET using DSRC. Range based CP can be categorized as radio based and non-radio based ranging where radio based range measurements can be observed using the radio signals emitted from DSRC transceivers. On the other hand, non-radio ranging can be derived from GNSS measurements such as C/A pseudoranges. This section will first present the radio range/range-rate based CP such as Time of Arrival (TOA), Round-Trip Time (RTT), Time Difference of Arrival (TDOA) and Received Signal Strength (RSS), range-rate derived from Centre Frequency Offset (CFO) and will proceed with non-radio range based CP. It is worth mentioning that CP can also be in the form non-range CP, where participating vehicles share satellite related information such as Doppler frequency shirt and carrier to noise ratio to speed up satellite acquisition time but is not discussed here as it is not in the scope of this paper. This section will also discuss some of the algorithms that are widely used in Wireless Sensor Network (WSN), which can directly be applied in VANET. 2.1 Radio Range/Range-Rate and non-radio range CP Techniques Time propagation measurements can be in two forms; the first technique, one-way propagation time measurement, allows for range estimation by measuring the time it takes from where the signal is transmitted to where it is being received. This requires accurate time synchronization of the transmitter and receiver, therefore, adds the complexity to the system. The second technique is the round-trip propagation time, where it measures the difference between the time when a signal is transmitted from a node to another and sent back to itself. Given that the two nodes are using the same clock, it removes the need for accurate time synchronization (Alam, 2012). Time difference of arrival technique allows for ranging estimation by measuring the time difference between the time when anchor nodes receive the transmitted signal from a non-anchor node, also known as multi-lateration. The mostly used technique of measuring TDOA is by using the generalize cross-correlation method, which can be obtained by integrating the lag product of two received signals for a particular time period. Once the time differences are obtained, one can compute the difference of angles and subsequently uses the known baselines between anchor nodes to compute their ranges to the non-anchor node. One of the drawbacks of TDOA is that severe effect of multipath can cause overlapping cross-correlation peaks which makes time difference estimation not possible (Mao et al., 2007). The RSS, which are readily available from most wireless devices, can be converted into ranges. This approach is attractive in WSN localization as it requires no additional hardware and wireless devices such as wireless access points are abundant, particularly in urban areas. The challenge of utilizing RSS is to firstly map the available access points, which is an impractical and laborious task. Secondly, it is hard to accurately model RSS as the signals are easily affected by reflection, scattering and diffraction which changes from one place to another (depending on wall thickness,

3 reflective surfaces present in the area, etc.) (Mao et al., 2007). Although radio ranging techniques discussed look promising, its viability using DSRC is questionable. For example, TOA requires complex time synchronization which is not readily supported by DSRC in that its base protocol, IEEE is only accurate to only micro seconds. In order to achieve acceptable ranging measurements for CP, timing accuracy in order of nanoseconds is required. On the other hand, although TDOA does not require time synchronization, it can only be realised using DSRC when two nodes are using the same bandwidth. This significantly reduces the capabilities of DSRC which is therefore not suitable for VANET. RSS by itself is also not suitable for CP in VANET due to its inaccurate ranging as demonstrated in (Alam, 2012). Range-rate measurements can be estimated based on Doppler shift or integrated carrier phase difference between vehicles. Relating to CP, it is less used when compared to range based due to the lower amount of location related information. The range-rate between vehicles can be calculated if the carrier frequencies of the transmitted and received signals are known. Due to the crystal clocks used in DSRC, its CFO of the received signal, which consists of Doppler shift is affected by clock drift. Using DSRC in particular, the CFO can be estimated with resolution of around 100 Hz for 5.9 GHz frequency. One of the advantages when using CFO based range-rate is that Doppler shift is not affected by multipath as much as range based techniques, which makes it more viable for CP. However, it also has a disadvantage in that it is only useful when relative mobility between vehicles are above the level of range-rates noise, which is not usually achievable when they are travelling in the same direction (Alam, 2012). Ranges in VANET can also be derived from sources that do not rely on radio signals of the communication transceivers (DSRC). For example, as will be elaborated in a later section, code based pseudoranges can be shared amongst the participating vehicles. This means that DSRC acts only as a medium to transfer data between vehicles which may include vehicles GNSS pseudoranges, positions and their variances. As shown in (Alam, Tabatabaei Balaei and Dempster, 2012), the estimated relative positioning using code based double difference can outperform DGPS in terms of accuracy. However, more study is needed to assess the viability of this technique when being employed in high multipath environments as it can severely affect the shared pseudoranges. 2.2 Algorithms for CP enablement Perhaps one of the most widely used algorithm in CP, the Kalman Filter (KF) is a recursive algorithm that uses a series of prediction and measurement update steps to obtain an optimal, in a minimum variance sense, estimate of the state vector. It is used in many applications involving localization and integrated systems, particularly in the field of navigation and tracking systems (Dellaert et al., 1999). The KF algorithm can be categorized into the prediction and update groups. Essentially, the prediction group describes how the state vector and its covariance propagate through time, based on the current state and assumed system model. Then, the update group updates the Kalman gain, state vector and system variance. The Kalman Gain, in a loose sense, weighs the process and measurements accordingly, taking account of their respective variances. Then, using the Kalman gain, the state vector is updated with new measurements. Finally, the system variance is updated, using both Kalman gain and the apriori variance (Hofmann-Wellenhof and Lichtenegger, 2001). The algorithm is then recursively applied to subsequent epochs. For non-linear systems, the Extended KF (EKF) is widely used where the process or measurement non-linear models are linearized before implementing the KF. Monte Carlo Localization (MCL), also known as particle filter (PF) is widely used for robot localization in an indoor environment where it uses fast sampling technique to represent the robot s belief. As presented in (Dellaert et al., 1999), the MCL algorithm is summarized as; starting with a prediction phase, a set of particles S k 1 is sampled. Then each set is applied through a motion model by sampling p(x k s i k 1, u k 1). This results in a new set of S k, which approximates a random sample from the predictive density p(x k Z k 1 ). Next, the update phase is applied where measurements z k are taken into account to weigh all of the sampling sets, which is given by m i k = p(z k s i k ). Then S k is computed by re-sampling from the weighted set. These two phases are repeated recursively for the subsequent steps. The MCL has some advantages over the other algorithms such as; unlike KF, MCL is able to represent multi-modal distributions, which is useful for self-localization and it is relatively easy to implement. The MCL algorithm is not only limited to robot localization, but extends to WSN as shown in (Wang and Zhu, 2007). Another algorithm that is recently introduced for WSN localization is the SPAWN algorithm. The SPAWN algorithm makes use of factor graphs (FG) and sum product algorithm (SPA) where the FG is a method of graphically represent a factorization of a Bayesian network while the SPA is a message passing algorithm for performing inference on the FG. Consider a WSN consisting a set of nodes and a set of anchors. Each node elaborates its information from the previous step from its last position estimation. Then it receives messages from visible anchors and neighbouring nodes. Using the new information, it updates its positional estimation and shares it with its neighbours. The messages shared among the nodes represent probability density function. This makes the approach a truly distributed algorithm which is highly suitable for CP (Wymeersch et al., 2009).

4 3. CURRENT AND FUTURE WORK This section presents the current work of the joint working group of FIG Commission 5 and IAG Commission 4, where most of the work has been focusing on the viability of using DSRC, using range-rates and non radio based ranging techniques as tools for enabling vehicular CP. The proposed CP techniques are mostly validated using Cramer Rao lower bound (CRLB) and root mean squared error (RMSE) obtained when compared to truth solutions. Due to the constraint of the paper length, not all of the current work are presented here. 3.1 CP based on range-rates Due to complexity and non accurate ranging achievable using techniques such as TOA, TDOA and RTT, Alam (2012) proposed on a CP system that uses CFO to derive range-rates between vehicles. In this work, GNSS position and derived inter vehicle range-rates are loosely integrated using the EKF. Due to the nature of Doppler shift, the CP is only considered for vehicles that are travelling in the opposite direction. The idea is that the target vehicle calculates the CFOs of the received signals from vehicles of opposite directions (neighbours), which are proportional to the range-rates between the target vehicle and its neighbours. The target vehicle then fuses its GNSS based position and velocity Doppler shifts of the received signals in order to improve its position estimates. From simulated experiments, position enhancement between 28% and 47% can be achieved which corresponds to 7.2 m to 5.3 m precision with CP, when compared to GNSS based positioning of 10 m. Yet another range-rate technique is proposed in Alam, Balaei and Dempster (2012) where it does not rely on GNSS and is able to provide instantaneous lane-level positioning. Unlike the previous work, this approach proposes a V-I CP, where two DSRC transceivers are placed on opposite sides of a road, which then broadcast their position and geometry of the lanes. Then, the passing vehicle uses these data, the CFO of the received signals, and odometer based speed to estimate its instantaneous position along the road and the lane. This technique is validated using both simulated and experimental data, and the results show that it is able to achieve a positioning standard deviation of less than 2 m and 0.5 m for when vehicles are travelling at low (< 20 km/h) and high speeds, respectively. 3.2 CP based on non-radio ranging Alam, Tabatabaei Balaei and Dempster (2012) have demonstrated how non-radio ranging CP technique can improve relative positioning by exchanging low level GNSS pseudoranges using DSRC. The study has shown that non-radio ranging between two moving vehicles can achieve better accuracy of relative positioning than differential GPS (DGPS). In this work, the pseudoranges are used to derive code based double differences which results in elimination of spatially correlated GNSS errors such as ionosphere, troposphere, satellite orbit and clock drifts effects. The proposed system provides elimination of infrastructure costs, as required by conventional systems such as DGPS. The derived double differences are tightly integrated which are then used to estimate the relative positioning between vehicles. The results of proposed technique show that its CRLB improved up to 30% when compared to that of DGPS. Similarly, its relative positioning RMSE also improved by 37% which concludes its superiority over the conventional technique. However, a shortcoming of the proposed technique is that it needs at least four satellites for it to be functional. Hence, it might not be viable in GNSS challenging environments such as in urban areas. In trying to improve the non-radio ranging CP technique, Hasnur-Rabiain et al. (2013) has demonstrated a similar approach with an additional sensor, a low cost MEMS based Inertial Navigation System (INS). Like the previous technique, the DSRC is used to transfer information between vehicles and pseudoranges are used to derive code based double differences. This work also uses the tight integration approach where derived double differences, pseudoranges and INS measurements are tightly integrated. The tightly integrated with INS approach delivers several benefits, one of which is the vehicles must only observe at least two common satellites for this technique to function. This system has shown vast improvement over conventional tightly coupled INS/GNSS during limited GNSS availability. In this work, the CP system is tested in three different scenarios; its performance with full GNSS availability, limited availability of three satellites and lastly, two satellites. To observe its performance further, the duration of the simulated GNSS partial outages is varied from 60, 180 to 300 seconds. Under full GNSS availability, the proposed system provides limited improvement but significantly reduces the positioning error when GNSS availability is limited. For example, up to 60% and 40% improvements over standalone INS/GNSS are reported when only three and two common satellites are available over 60 seconds, respectively. In other words, the system is able to maintain RMSE of below 2.5 m compared to INS/GNSS RMSE of 5.7 m. Even after 300 seconds of partial availability, the proposed system continuously outperformed INS/GNSS by 50% and 25% when three and two satellites are available, respectively. 3.3 Algorithm and Improved Communication Apart from the conventional approach of using EKF, Efatmaneshnik et al. (2012) has proposed a non classic fast multidimensional scaling filter (MDSF) for vehicular CP. The proposed algorithm is based on the MDS solution

5 covariance estimation technique and the maximum likelihood (ML) filter. Compared to the conventional MDS, it is computationally more effective particularly when map matching is infused in the system, considerably reducing the number of iterations and convergence time which is important in the topologically fast changing vehicular environment. Using simulated data, the MDSF is compared against EKF under light, medium and heavy traffic conditions. It is assumed that the vehicles are able to provide inter-ranges with a standard deviation between 1 to 10m. The results show that the MDSF outperformed EKF by 0.2 m. Furthermore, the study has shown that when GPS drop outs are simulated, both MDSF and EKF managed to continuously provide positioning solutions beyond expectations. This shows how CP could benefit VANET particularly in GNSS denied environments. V-V and V-I communications via DSRC in a dense network can a pose severe challenge in trying to successfully exchange information packages. Frequent exchange of large amounts of range information required by existing CP schemes increases the packet collision rate of the vehicular network which reduces the effectiveness of the CP. To address this issue, a simple, easily deployable protocol improvements in terms of utilizing as much range information as possible was proposed in (Yao et al., 2011). The protocol improvements also reduces range broadcasts by piggybacking, compressing range information, tuning the broadcast frequency, and combining multiple packets using network coding. The results demonstrate that, even under dense traffic conditions, these protocol improvements could achieve a twofold reduction in packet loss rates and increase the positioning accuracy of CP by 40% (Yao et al., 2011). 3.4 Dataset Collection In May 2012, a series of field experiments revolving around the concept of collaborative positioning and navigation have been performed in Nottingham University, UK, which details can be found in (Kealy et al., 2012). In total, the field experiments lasted for four days where different scenarios were considered; indoor, outdoor and a combination of both. Several platforms including a train on a building roof, a foot tracker, a personal navigator and mobile mapping vans were deployed. They were fitted with various sensors such as geodetic and low-cost high-sensitivity GNSS receivers, tactical grade INS, low cost MEMS-based INS, magnetometers, barometric pressure, step sensors, as well as image sensors, such as digital cameras and Flash LiDAR, and ultra-wide band (UWB) receivers. The experiment that covers land mobile CP in particular were conducted using two vehicles, each fitted with high grade INS, MEMS based INS, geodetic GNSS receivers and DSRC transceivers. The vehicles then were driven on road sections near the University of Nottingham. The experiment lasted for two hours where it was separated into several loops. For each loop, the power parameters of the DSRCs were reconfigured to observe their resulting RSS and CFO observation noise. The trajectory of the test track has both low and medium dynamics in terms of velocity and turning profiles, depicting typical land based vehicle dynamics. For example, velocity along the trajectory varied, reaching up to 90 km/h and has several slow/slight and relatively fast turns. The satellite availability for this particular test is about 95%, where several outages were experienced while driving under bridges. 3.5 Future Work The initiative of the joint IAG and FIG working groups to improve CP will take on various aspects, mainly, improving DSRC measurements or observations, using smarter algorithms with low computational load, incorporating other types of sensors and adding constraints to hamper the positional growth of CP in VANET. As discussed in earlier section, ranging using DSRC are not feasible for vehicular applications. However, the range-rates has shown promising results but is limited to the motion of participating vehicles in that they have to be travelling in the opposite direction. Non-radio ranging CP on the other hand is accurate but requires vehicles observe common satellites simultaneously. This might not always be feasible particularly in urban environments. Hence, a hybrid of radio ranging and non-radio ranging is proposed to further improve the feasibility of DSRC in VANET. ITS applications such as collision avoidance requires lane level navigation or sub-metre accuracy, therefore vehicular communication cannot be the only technique to achieve such accuracy. It has been shown that the inclusion of INS has significantly improved the performance of CP. Numerous studies which utilized other types of sensors such as short range radar and laser ranging have also shown considerable improvement of positioning solutions (Garcia et al., 2008). Hence, it is the interest of this group to use other ranging sensors, readily applicable in vehicles to further improve CP. However, great care must be taken when incorporates more sensors as DSRC has a limited bandwidth which limits the amount of data that are transferable. Hence, this group will also investigate the feasibility of how much information can be shared between vehicles VANET. Rather than using standard EKF approach, the research will test on other estimation techniques such as using Unscented KF, Particle filter, SPAWN, etc. to further improve CP technique. Also, alternative dynamic model will be developed specifically to accommodate land based motion which would improve positioning output. Furthermore, map-matching techniques including deterministic, Bayesian, fuzzy logic and set-membership methods for CP will be tested as they could be beneficial in VANET, where land based vehicles are often constrained by road networks and geometry.

6 4. CONCLUSION This paper has presented the theory of CP specifically applicable for land mobile applications. It has described CP based ranging, range-rating and non-radio ranging techniques and their viability using DSRC. WSN algorithms such as KF, MCL and SPAWN, which can be applied directly for CP in VANET have also been discussed in this paper. Current initiatives undertaken by a joint working group, consisting of FIG Commission 5 and IAG Commission 4 have been presented where some of the works elaborated in this paper include using range-rates and non-radio ranges for CP. The group has also worked on using different algorithms and improving DSRC communication methods to enhance CP in terms of positioning accuracy and availability. Finally, the paper has discussed some of the challenges to improve CP, which will be the focus of the group s future work. REFERENCES Alam, N. (2012), Vehicular Positioning Enhancement using DSRC, PhD thesis, School of Surveying and Spatial Information Science, The University of New South Wales. Alam, N., Balaei, A. and Dempster, A. (2012), An Instantaneous Lane-Level Positioning Using DSRC Carrier Frequency Offset, Intelligent Transportation Systems, IEEE Transactions on 13(4), Alam, N., Tabatabaei Balaei, A. and Dempster, A. G. (2012), Relative Positioning Enhancement in VANETs: A Tight Integration Approach, Intelligent Transportation Systems, IEEE Transactions on PP(99), 1 9. Dellaert, F., Fox, D., Burgard, W. and Thrun, S. (1999), Monte Carlo localization for mobile robots, in Robotics and Automation, Proceedings IEEE International Conference on, Vol. 2, pp vol.2. Efatmaneshnik, M., Alam, N., Kealy, A. and Dempster, A. G. (2012), A Fast Multidimensional Scaling Filter for Vehicular Cooperative Positioning, The Journal of Navigation 65(02), URL: Garcia, R., Aycard, O., Vu, T.-D. and Ahrholdt, M. (2008), High level sensor data fusion for automotive applications using occupancy grids, in Control, Automation, Robotics and Vision, ICARCV th International Conference on, pp Hasnur-Rabiain, A., Kealy, A., Alam, N., Dempster, A., Toth, C., Brzezinska, D., Gikas, V., Danezis, C. and Retscher, G. (2013), Cooperative Positioning using GPS, Low-cost INS and Dedicated Short Range Communications, in Proceedings of the PACIFIC Positioning, Navigation and Timing of the Institute of Navigation (ION PNT 2013). Hofmann-Wellenhof, B. and Lichtenegger, H. (2001), GPS Theory and Practice, fifth edn, Springer, New York. Kealy, A., Retscher, G., Alam, N., Hasnur-Rabiain, A., Toth, C., Grejner-Brzezinska, D., Moore, T., Chris Hill, C., Gikas, V., Chris Hide, C., Danezis, C., Bonenberg, L. and Roberts, G. W. (2012), Collaborative Navigation with Ground Vehicles and Personal Navigators, in 2012 International Conference on Indoor Positioning and Indoor Navigation. Mao, G., Fidan, B. and Anderson, B. D. (2007), Wireless sensor network localization techniques, Computer Networks 51(10), URL: Parker, R. and Valaee, S. (2007), Vehicular Node Localization Using Received-Signal-Strength Indicator, Vehicular Technology, IEEE Transactions on 56(6), Wang, W. and Zhu, Q. (2007), Varying the Sample Number for Monte Carlo Localization in Mobile Sensor Networks, in Computer and Computational Sciences, IMSCCS Second International Multi-Symposiums on, pp Wymeersch, H., Lien, J. and Win, M. (2009), Cooperative Localization in Wireless Networks, Proceedings of the IEEE 97(2), Yao, J., Balaei, A., Hassan, M., Alam, N. and Dempster, A. (2011), Improving Cooperative Positioning for Vehicular Networks, IEEE Transactions on Vehicular Technology 60(6), Zador, P., Krawchuk, S. and Voas, R. (2000), Final Report Automotive Collision Avoidance System (ACAS) Program, Technical report, National Highway Traffic Safety Administration, U.S.