Physical modeling of environment detection sensors, based on GPU-supported shader technology

Abstract

Advanced driver assistance systems are mainly based on interpreting traffic situations using the sensor data they get. During the development and trial of the algorithms, it is crucial to provide data sets that are as realistic as possible. There are countless different situations in traffic; to reproduce each of them by doing a test ride is very costly. Therefore the simulation of these situations with virtual sensor models can save effort, time and even replace the test ride in some situations. Flexibility can be achieved by deterministic models, which need very high computation power which complicates the real-time simulation. Sensor physics can be broken down to sending of rays or waves which are reflected by their environment and measured afterwards. The ideal hardware for this kind of real-time simulation already exits – the graphics processing unit (GPU). GPUs are designed for high-grade parallel processing and they include methods for calculating reflections through so called shaders. The challenges are identifying the analogies between shader based light calculation and sensor physics, as well as using them in order to develop a realistic model that can be run in real-time. Ultrasonic sensors used in parking aid systems will benefit most from deterministic simulation. Thus, this work concentrates on these kinds of sensor.