Automotive Camera Simulation for Testing ADAS and Active Safety Systems

Abstract

In recent years the automotive industry has increasingly used virtual camera sensor simulation to test advanced driver assistance systems and active safety systems. These simulations need to be as close to the real world as possible to be representative for real world scenarios. An accurate light representation in the render engine is of paramount importance to the simulation's realism. Also weather effects have a large impact on the performance of vision based controlling systems. In this thesis we investigate both subjects. We construct a custom color space based on the sensitivity curves of an automotive camera to better approximate the light in a render engine that is only able to process trichromatic color vectors. Our results show however that using single wavelength primaries is not the solution for doing this. Furthermore, we show how water drops on the windscreen can be simulated with both our a ray tracer implementation and a direct mapping method. We demonstrate that internal reflections become more important for the end result when the water drops have a large contact angle or when they are at an inclination angle with respect to the lens of the camera. The direct mapping method is not able to simulate these reflections.