MODEL-BASED VEHICLE TRACKING FROM IMAGE SEQUENCES WITH AN APPLICATIONTO ROAD SURVEILLANCE

A model-based approach to vehicle tracking is proposed and applied to a highway traffic surveillance problem, which is motivated by current research in intelligent transportation systems. Systems for traffic ma nagement and traveler information services require accurate and wide-a rea estimates of vehicle velocity and traffic spatial and temporal den sities. A detection and tracking algorithm is developed that achieves good performance with complexity low enough for real-time implementati on using inexpensive microprocessors. Detection thresholds are compute d based on a statistical model for vehicle and background, and the the oretical detector performance is derived. The tracking algorithm filte rs position estimates from the detection algorithm using a simple vehi cle dynamic model and the Kalman filter. Data association is accomplis hed with a nearest neighbor filter coupled with a lane-change handling logic. (C) 1996 Society of Photo-Optical Instrumentation Engineers.