Font Size: a A A

Study On The Tracffic Event Detection Based On Video Tracking

Posted on:2011-01-24Degree:MasterType:Thesis
Country:ChinaCandidate:F DongFull Text:PDF
GTID:2178360308460656Subject:Traffic Information Engineering & Control
Abstract/Summary:PDF Full Text Request
This paper analyses the major problems of automatic detection techniques of traffic events, and studies the key technique of traffic events detection system based on video tracking, then proposes a traffic events detection method based on the analysis of target trajectories. With traffic video images captured by cameras, binary segmentation, target tracking, trajectory extraction and other events detection relevant techniques are analyzed and studied.During the procedure of binary segmentation, an improved multi-frame image average method is applied, and a block-based binarization method is proposed, which divides an image into several unoverlapped blocks and binarizes and segments the image block by block by computing the average and the variance of the pixels in each sub-block.A tracking method using block matching, which is based on target corners, is used in this paper.A corner detection algorithm based on the Sobel edge detection operator is used to extract target corners, and target trajectories are obtained by using a full-search matching method with the block in which target corners exist as the template. A pseudo-trajectory processing method based on the relationship between chord lengths and arc lengths is proposed to obtain reasonable trajectories.A method for obtaining the target moving directions is presented after studying and analyzing target trajectories.According to the characteristics of traffic event trajectories and vehicle speeds, a detection method for low vehicle speed, speeding, retrogradation, lane change and other traffic event is presented. And a detection method based on speeds and bounding rectangular areas is proposed for pedestrian and parking events detection.The presented algorithm was tested in different traffic scenarios. Experiment results show that this algorithm has great adaptability to occlusion and adhesion on targets, and its relatively low computation satisfies the requirement for real-time tracking. The proposed events detection algorithm can correctly detect abnormal vehicle behaviors and realize the automatic incident alarm as well.
Keywords/Search Tags:traffic event detection, block matching, corner extraction, trajectory, binarization, background extraction
PDF Full Text Request
Related items