Evaluation Method for Automotive Stereo-Vision Systems

Safe vehicle guidance under human or computer control requires a thorough understanding of the traversed environment. Consequently if perception systems are to be introduced into mass market vehicles as part of driving assistance systems, their proper operation throughout the vehicle working life is needed. Onboard stereo-vision systems can provide rich information in terms of range, feature recognition, etc., hence the interest by car OEMs. System performance depends on multiple factors like light conditions, algorithms and the mechanical apparatus. Due to inaccuracies produced by changes in the system physical properties due to vibrations, misalignment of fixtures, etc. through the vehicle operational life a reduction in performance will occur. In this paper, an evaluation framework to estimate the performance of a vehicle onboard stereo-vision system in terms of 3D measurements and re-projection errors is presented. The approach considers changes that might occur in the system during the vehicle working life. It includes means to evaluate the self-calibration process often used to correct the effects of physical changes in the stereo-vision system. The results provide key information for the design and geometrical specification of automotive stereo-vision systems. As the potential physical changes in the geometric configuration of the camera-pair over the vehicle life time are difficult to predict, it was necessary to simulate them to generate families of errors that these might trigger on the system performance. The results of the observations and analysis are included; these should assit designers to define the constraints that needed for the layout of cameras as part of the design process.