A SYSTEM ARCHITECTURE FOR VISUAL TRAFFIC SURVEILLANCE
Nhc Yung · 1998
In this paper, a VTS architecture is proposed, which incorporates background estimation, lane detection, occlusion detection, feature extraction and parameter estimation into a framework of vehicle extraction, modeling, motion estimation and tracking. In principle, the background estimation uncovers the stationary background, which can be used for extracting the moving vehicles. The lane detection produces centerlines that can simplify the modeling and motion estimation. The occlusion detection reduces ambiguity arising from motion estimation. The feature extraction allows identification and labeling of individual vehicles whereas the parameter estimation retrieves information such as flow rate and saturation rate. Extensive testing of the proposed architecture on real traffic sequences shows that it is indeed capable of dealing with occlusion, tracking vehicles and estimating travel parameters in an accurate and robust manner. For the covering abstract see IRRD E102946.