An interactive model based vision system for vehicle tracking

An architecture for an interactive model-based vision system and its application for vehicle tracking are described. A human specifies a limited amount of information which establishes a context for autonomous interpretation of images obtained by a telerobot. Object models are described by constraints specifying necessary geometrical properties and relationships between objects. The use of constraints allows for flexible object instantiation. A user can indicate a vehicle and this directs perceptual processing routines to determine the corresponding local surface orientation and roads, or the user can instantiate a road segment to direct the extraction and tracking of vehicles. A processing example based upon implemented components is given, and a brief discussion of future work is provided