Diss. Göteborg : Chalmers University of Technology. Department of Signals and Systems. Signal Processing Group, 2010

This thesis studies the problem of tracking in the setting of an automotive safety system. In particular, it considers the problem of estimating the surrounding traffic situation using observations from radar sensors. In this context, we develop two accurate radar sensor models and vehicle tracking algorithms, when multiple measurement can be obtained from each object. The first model describes the radar return from a vehicle as originating from a known set of point features, whereas the second approach jointly estimates the position of the reflecting point features and the position of the extended object. Both models incorporate novel approaches for describing the limited resolution of the sensor and the resulting tracking frameworks effectively exploit the information in all the detections from the vehicle. Additionally, we investigate the use of radar measurements in a probability hypothesis density (PHD) framework for constructing maps over the stationary objects around a vehicle. By proposing new data clustering and merging methods we manage to exploit the inherent structure in the map. The efficiency of the PHD framework both in the measurement update and in the representation of the map is thereby improved considerably. Besides models for accurately describing the measurements, we also propose a new vehicle motion model that describes the driver as an optimal controller with preferences (described by a parameter vector) which are tracked over time. The proposed framework enables more accurate predictions and offers a formal treatment of the model uncertainties. Finally, we present a modular sensor data fusion functional architecture, tailored for development of automotive safety systems. The ambition with this paper is to illustrate how the other findings in the thesis can be implemented in practice.