Scene Based Safety Functions for Pedestrian Detection Systems 2013-26-0020

The protection of pedestrians from injuries by accidental collision is a primary focus of the automotive industry and of government legislation [1]. In this area, scientists and developers are faced with a multitude of requirements. Complex scenes are to be analyzed. The wide spectrum of where pedestrians and cyclists appear on the road, weather, and light conditions are just examples. Data fusion of raw or preprocessed signals for several sensors (cameras, radar, lidar, ultrasonic) need to be considered as well. Accordingly, algorithms are very complex. When moving from prototypic environments to embedded systems, additional constraints must be considered. Limited system resources drive the need to simplify and optimize for technical and economic reasons. With all these constraints, how can the safety functions be safe-guarded? This submission considers scene-based methods for the development of vehicle functions from prototype to series production focusing on functional safety. Starting points are modeled use cases which can be evaluated by simulation and variation of internal parameters as well as external influences, like velocity, course, likelihoods, measurement errors and latencies. Based on these methods, the approach of scene based specifications is discussed. Scene based test catalogs are also considered. These test catalogs can consist of synthetic (simulated) test cases and recorded sequences. The effect of changes during a development project can be more reliably evaluated with stable and reproducible test cases. The article presents typical sample usage scenarios and discusses the results achieved so far.