A Novel Traffic Stream Detection Method Based on Inverse Perspective Mapping

Abstract Traffic stream detection in intersection is an important part of Intelligent Transportation System (ITS). The traditional image-based detection methods are vulnerable to background noise and the perspective effect of stationary camera, thus we propose a novel traffic stream detection method based on inverse perspective mapping (IPM). The method firstly eliminates the geometric distortion of sequence image using IPM transform, and extracts the marking lines in lane area by introducing in geometric constraints of structural road. On this basis, we extract the contours of vehicles queue using background difference method, and give two kinds of metrics to measure traffic stream, that is vehicles contour area based method and vehicles queue length based method. Experiment results show that the proposed algorithm is effective and robust.