Shadow-based vehicle model refinement and tracking in advanced automotive driver assistance systems

Abstract

Vision-based automotive driver assistance systems have to cope with complex outdoor illumination conditions. Many applications for vehicle detection and tracking consider shadows caused by the sunlight as distracting effects and try to computationally compensate or disregard them. In this paper we suggest a shape-from-shadow approach which uses cast shadows as additional supporting information for vehicle model refinement and tracking. To take the position of the sun into account we only use sensor systems that mid-range vehicles are normally equipped with. Analysing shadows is a suitable method to support rear-view based vehicle tracking. A vehicle's shadow turned out to be a strong feature since it is possible to track a vehicle solely based on its shadow. Another benefit is that we are able to set up and refine three-dimensional shape models of vehicles driving ahead in the same lane. In selected situations it is possible to visually track two vehicles driving on the same lane one after another.