A Concept of Camera Test-Bench for Testing Camera Based Advanced Driver Assistance Systems

Abstract

About 1.5 million traffic accidents occur every year in European traffic; most of them are results of careless driving behavior, for example unintentional lane departure. Studies show that these accidents could often be avoided, if the driver would react only half a second earlier. Therefore, advanced driver assistance systems (ADAS) and especially the camera based ADAS increasingly gain importance in the automobile industry because of their diverse functions, for example the lane keeping assistance. Their development must contain an excessive and accurate testing process, in order to contribute to a reliable result in the common use. The importance of the testing methods and environments is growing due to several reasons, for example the need for more flexibility during testing and cost-reduction at the same time. Those requirements can be fully fulfilled by the use of virtual environments. To simplify the testing process, the virtual environment does not need to be as realistic as possible. It only has to provide necessary features for the camera system. Due to different detection algorithms for diverse assistance functions, the virtual environment needs determined characteristics, which are crucial for further precise analyses. This paper presents a concept to create a camera test-bench for testing of camera based ADAS, by adapting an existing virtual environment for the camera system, identifying and enhancing the important features for the testing detection function while enable a flexible and low-cost testing environment. The testing ADAS shall record the virtual driving scenes and deliver the same results as in real environments. First, possible existing approaches of testing processes shall be presented. In this paper, the concept is to shown by the example of road detection. The focus lies on the lane markings and the ground. While the separation lines between diverse objects are clearly visible in the virtual environment, the real ones often differ due to road condition or lighting intensity. The differences must be considered and corrected by creating a shader which adjusts the scenes, before the virtual representations can be applied for testing camera based ADAS. Furthermore, for testing other functions, the shader shall be expanded and contain more relevant feature parameters for different test functions. All in all, as much as possible camera based ADAS shall be tested with a virtual environment and the tests shall help accelerating the development time and improving the quality of camera based ADAS.