This paper deals with the problem of estimating the vehicle surroundings (lane geometry and the position of other vehicles), which is needed for intelligent automotive systems, such as adaptive cruise control, collision avoidance and lane guidance. This results in a nonlinear estimation problem. For automotive tracking systems, these problems are traditionally handled using the extended Kalman filter. In this paper we describe the application of the marginalized particle filter to this problem. Studies using both synthetic and authentic data show that the marginalized particle filter can in fact give better performance than the extended Kalman filter. However, the computational load is higher.