Abstract
The introduction of perception systems as part of active safety systems in mass market vehicles requires systems capable of operating all the time and providing the correct information. Their validation could be complex and expensive. This paper presents the development of a virtual environment in which 3D road scenarios can be modelled including radiation models that reflect changes in light or temperature. The resulting synthetic images are used to excite sensor models, emulating their response. For example images from a camera pair can be used to generate disparity maps and evaluate stereovision algorithms. It is thus possible to experiment virtually multiple configurations in identical conditions.
By using 3D vehicle models, the effects of sensor layouts can be examined. The availability of vehicle dynamics models means that dynamics effects can be incorporated. Therefore synthetic images representing scenes observed by the sensors as the vehicle moves can be generated. Accelerations and synthetic images can be used to run localisation algorithms that incorporate proprioceptive and exteroceptive sensors.
The paper presents this design platform, the scenario creation, irradiance and sensor models. It includes a case study that comprises the design of a perception system for lane detection. The paper centres on the manner in which this tool is used as a virtual development platform that enables designers to test multiple-hypothesis prior to their implementation in physical test-rigs whilst at the same time expanding our understanding of the perception process.
Keywords:perception systems, synthetic images, simulation, sensor- vehicle models