A New Decentralized Bayesian Approach for Cooperative Vehicle Localization based on fusion of GPS and Inter-vehicle Distance Measurements Mohsen Rohani, Denis Gingras Electrical and Computer Engineering Department Université de Sherbrooke Sherbrooke, Canada mohsen.rohani@usherbrooke.ca, denis.gingras@usherbrooke.ca Vincent Vigneron Université d’Evry Evry, France vincent.vigneron@ufrst.univ-evry.fr Dominique Gruyer IFSTTAR- IM - LIVIC Satory, France dominique.gruyer@ifsttar.fr Abstract— Embedded intelligence in vehicular applications is becoming of great interest since the last two decades. The significant growth of sensing, communication and computing capabilities over the recent years has opened new fields of applications, such as ADAS and active safety systems, and has brought the ability of exchanging information between vehicles. In this paper, a new method for improving vehicle positioning is proposed. This method is a decentralized method based on sharing GPS data and inter-vehicular distance measurements within a cluster of vehicles. A Bayesian approach is used to fuse the GPS data and inter-vehicular distances. In order to investigate the performance of this new approach on vehicle localization, a Kalman filter has been employed to incorporate the dynamics of the vehicle. The effect of this method on the reduction of the localization uncertainty, over-convergence issues and identification of the vehicles are also discussed in this paper. Keywords— Cooperative Vehicle Localization; GPS; Inter- vehicle Distance; VANET; Bayesian I. INTRODUCTION Accurate and reliable vehicle localization is a key component of numerous automotive and Intelligent Transportation System (ITS) applications, including active vehicle safety systems, real time estimation of traffic conditions, and high occupancy tolling. Various safety critical vehicle applications in particular, such as collision avoidance or mitigation, lane change management or emergency braking assistance systems, rely principally on the accurate and reliable knowledge of vehicles’ positioning within given vicinity. Distributed algorithms as proposed in [1] and [2] have underlined a recent and important interest for the collaborative localization. Since the number and type of sensors used in vehicular applications increases, it is essential to find ways to better analyze and extract useful data from these sensors and share them between vehicles when it is relevant. Sensors which are used in localization can be divided in two categories: proprioceptive and exteroceptive sensors. Proprioceptive sensors are those which can provide information about the vehicle dynamic states like position, velocity and acceleration (GPS, accelerometer, gyroscope etc.). Exteroceptive sensors provide information about the states of the environment (video camera, lidar, etc.). GPS is one of the most common positioning devices being used in vehicle Range Sensor Satellites’ Signals Proposed Method VANET Data Signals Received from other Vehicles GPS Receiver Cooperative Position Estimates Data Signals sent to other Vehicles Inside a Vehicle Kalman Gain Update Estimate Update Covariance Project to K+1 Kalman Filter New Estimated Position & Covariance Figure 1 - Schematic of the system, the input data (blue), the proposed method (green), the Kalman filter (orange) and the new Position Estimation