A vehicular filter suitable for co-operative automotive safety applications

Research in the automotive field is currently focusing on exploiting the co-operation among vehicles to enhance road safety, efficiency and comfort. In vehicular ad hoc networks, especially in safety related applications, accuracy of the information exchanged through the wireless network, such as position and velocity of each vehicle, is crucial. For this reason a robust and accurate vehicular filter which also reduces measurements´ noise is needed. The extended Kalman filter is used as standard technique for performing recursive nonlinear estimation. However, it provides only an approximation to optimal nonlinear estimation. In order to avoid sub-optimal performance an enhanced unscented Kalman filter is used. The enhancement is that a combination of different motion models is exploited, leading to a more accurate modeling of the vehicle´s movement. For this combination the curvature of the road from a digital map and the yaw rate and acceleration values from the vehicle are taken into consideration. The performance of the proposed unscented filter is proven to be superior compared to the performance of both the extended Kalman filter and the simple unscented filter which takes into account only a specific motion model.