Video-Based Vehicle Detection Scheme in Complex Traffic Scene at Urban Intersection

Abstract

Abstract —A new and rapid scheme for detecting vehicles based on video by single camera at urban intersection is proposed. Firstly a rapid judgment algorithm of vehicle direction is introduced, with motion object segmentation by means of frame-difference method and morphology. It extracts the uniqueinterested region from multiple objective through the modified region-growth algorithm E gets motion direction by 16 adjacent blocks matching using MAD criterion. Secondly, a new scheme is designed associating the above algorithm with virtual test linemethod in single direction. Moreover, a video detection system is developed on VC++ and DirectShow platform. Experimental results show that the accuracy rate of vehicle direction judgment is over 94% and the whole system can realize traffic statistics effectively. Keywords-vehicle detection, vehicle direction judgment, a single video camera, motion object segmentation, complex traffic scene I. I NTRODUCTION Intelligent Transportation System is an effective solution of contradiction between the fast development of the automotive industry and road capacity. Computer vision technology has a promising prospect in the transport system. Video-based vehicle detection is the key technology in Intelligent Traffic Monitoring System[1-2]. Different from some simple scenes (e.g. highway), there exist lots of difficult problems of video vehicle detection in complex traffic scenes at the urban intersection [3-4]. The difficulties are caused by some reasons such as light condition, too small moving target, camera resolution, camera vibration, visual effect of ‘near big’ and ‘far small’ etc. (Fig.1). Movement shadow and vehicle shelter are the most two important factors to the final accuracy rate. The former, the adhesion would bring errors in the ditection of neighboring vehicles, while the latter, vehicle overlap would lead into leak detection. In addition, the urban intersection might have different