Environmental perception with self-diagnosis for advanced driver assistance systems

Today, Advanced Driver Assistance Systems (ADAS) represent an increasing contribution to active road safety and driving comfort. Their task is to help the driver to avoid accidents or at least to minimize their consequences. Depending on their objectives they can support the driver's decisions by providing additional information (e.g. Park Distance Control, Night Vision, Traffic Sign Recognition) or even directly influence the driving process (e.g. Park Assist, Adaptive Cruise Control, Lane Assist). Since the absolute number of ADAS sensors in vehicles is permanently increasing, further development of existing sensor data processing mechanisms is required to ensure a robust functionality and (eventual) timely detection of system limits (e.g. caused by sensor misalignment, bad weather conditions, pollution or aging). Therefore, continuous knowledge of the quality of the environmental perception is of significant importance.

In this thesis, general probabilistic approach to multi-sensorial environmental perception of ADAS is presented. This approach incorporates sensor data fusion with self-diagnosis capability and maneuver level intent estimation of detected objects. Thus, the quality of environmental perception can be continuously monitored and the intents of the traffic participants can be predicted. The resulting probabilities are uniform and consistent basis and reflect the reliability of the results. This knowledge is an important prerequisite for the development of future complex and robust Advanced Driver Assistance Systems.

The developed concepts have been used and approved in project "Integrated Lateral Assistance", a subproject of research initiative AKTIV (abbreviation for "Adaptive and Cooperative Technologies for the Intelligent Traffic") supported by the German Federal Ministry of Economics and Technology.