The problem of automatically detecting a leading vehicle in theprocess of starting off and deriving a model for further tracking isaddressed. Very little knowledge about the appearance of the vehicle isutilized. Motion information is the main cue for selecting imagestructures belonging to the leading vehicle. An edge model is derivedfrom two independent, complementary motion-estimation processes thatcross-validate each other. The suitability of the aggregated model forrecognition and the precision of the motion estimation process aredemonstrated by means of an image sequence consisting of 130 TV images
展开▼