Research On Multi-Vehicle Tracking Based Urban Traffic Illegal Parking Detection

With the rapid development of economy and increase of motor vehicles, deaths caused by traffic accidents have increased dramatically as well as economic losses. Various kinds of traffic accidents have become serious problems which affect urban traffic safety, the main cause for traffic accident is that some drivers do not obey traffic rules. For traffic image contains the most comprehensive traffic information, video image processing technology has been widely applied in traffic violation behavior surveillance. Illegal parking is one of the most common vehicle violation behaviors, which may lead to serious traffic chaos and traffic accident. Video based illegal parking vehicle automatic capture system could monitor illegal parking vehicles on the roads timely, obtain illegal evidence, make drivers obey traffic rules consciously and reduce traffic accidents and traffic jam. This dissertation researches multi-vehicle detection and tracking algorithm used for traffic video monitoring. The main work of this dissertation includes:1. Various dynamic background modeling methods are introduced. Aiming at urban illegal parking detection, this dissertation analyzes and improves Gaussian mixture background model, updates background and eliminates the foreground noises combining with the correlations among adjacent pixels. The experimental results finally show that this method has good performance.2. Some object detection methods (e.g. two consecutive frames subtraction, optical flow and background subtraction) are discussed. Background subtraction technology is adopted to detect vehicle objects. In allusion to shadow influence, this dissertation compares the widely used shadow suppression methods, and presents a kind of HSV color space based detection lines subtraction shadow segmentation algorithm.3. Vehicle trajectory is predicted by Kalman filter and a real-time object tracking state judgment is achieved by using the feature matching matrix. Meanwhile, motor vehicle's illegal parking behavior is recognized according to the average and mean square error of tracking trajectory.4. An intelligent system of video-based illegal parking vehicle detection and automatic capture is implemented.