Simulation and evaluation of sensor characteristics in vision based advanced driver assistance systems

This paper presents a method for the simulation of images in the scope of virtual camera prototypes under the constraint of color correctness. This is a first step to gain a complete simulatable camera model that can be used to generate synthetic images using real data. Each real camera system has its own color processing characteristics. Real images recorded with a reference camera model can be computationally simulated as if they have been recorded with another real or virtual camera. The resulting images are transformed to underly the color characteristics of the targeted virtual camera system. Our approach can be used at the design phase of vision-based advance driver assistance systems to verify the exact behaviour under varying optical properties of the optical system and to test and evaluate the overall robustness of the system when color processing changes. It can as well lead to a decision basis for the selection of the hardware to be used. In this paper, we show how cameras can be calibrated and in a second step we evaluate the simulation errors. Finally, we apply our simulation to a traffic sign recognition algorithm and evaluate its behaviour in relation to ground truth data.