Post processing of laser scanner measurements for testing advanced driver assistance systems

Testing of advanced driver assistance systems demands a highly accurate representation of the vehicle's environment, e.g. obtained by laser scanner sensors. In contrast to typical on-line assistance functions, the purpose of generating reference data allows full batch processing of the raw sensor measurements. Therefore, object tracking algorithms can make use of both past and future measurement information. Hence, ambiguities during object detection and data association may be resolved in a principled manner. Although many works consider the tracking of vehicles using LIDAR1 sensors, the specifics of batch processing have not been investigated yet. This paper proposes novel non-causal approaches to detecting and tracking vehicles in 2-D scan point clouds given by an automotive LIDAR sensor. Evaluations based on experimental data show how post processing can improve the trade-off between the suppression of clutter measurements and prompt detection as well as the estimation accuracy.