Towards autonomous convoy driving - recognizing the starting vehicle in front

This contribution addresses the problem of automatically detecting a leading vehicle in the process of starting off and to derive a model for further tracking. Very little knowledge about the appearance of the vehicle is utilized. Motion information is the main cue for selecting image structures belonging to the leading vehicle. Further an edge model is derived from two independent complementary motion estimation processes cross-validating each other. The suitability of the aggregated model for recognition and the precision of motion estimation process is demonstrated by means of an image sequence consisting of 130 TV-images.