Improving dependability of vision-based advanced driver assistance systems using navigation data and checkpoint recognition

Abstract

Advanced Driver Assistance Systems (ADAS), like adaptive cruise control, collision avoidance, and, ultimately, autonomous driving are increasingly evolving into safety-critical systems. These ADAS frequently rely on proper function of Computer-Vision Systems (CVS), which is hard to assess in a timely manner, due to their sensitivity to the variety of illumination conditions (e.g. weather conditions, sun brightness). On the other hand, self-awareness information is available in the vehicle, such as maps and localization data (e.g. GPS). This paper studies how the combination of diverse environmental information can improve the overall vision-based ADAS reliability. To this extent we present a concept of a Computer-Vision Monitor (CVM) that identifies predefined landmarks in the vehicles surrounding, based on digital maps and localization data, and that checks whether the CVS correctly identifies said landmarks. We formalize and assess the reliability improvement of our solution by means of a fault-tree analysis.