Research On Vehicle Detection And Tracking Algorithm Based On Video

With the development of society and acceleration of urbanization, traffic congestion has become one of the tough problems that all the countries around the world are faced with. In order to optimize the traffic management, ITS is becoming a hot research subject at home and abroad in recent years. As a momentous part of Traffic Monitoring System, thus the detection and tracking technology of moving vehicles is an important research content in ITS. On the base of using video surveillance to detect and track the target vehicles, this thesis elaborates a system detecting and tracking multiple targets of vehicles at a time.The main content of this thesis consists of three modules:Detection Module, Trigger Module and Tracking Module. Detection Module studies the detecting algorithm of main moving targets at the present stage on the base of video surveillance, comparing and analyzing the implementing of optical flow method, inter-frame difference method, background subtraction method and etc. It also studies the background modeling in Single Gaussian Model and then puts forward an "anti deception system" to avoid misjudgment on dots of foreground. This system prevent phenomena of trailing and " unshell" efficiently. In the same time, this thesis also puts forward a modeling method based on modified Single Gaussian Model to collect information of the foreground of vehicles, which avoids misjudgment of foreground caused by sudden change of light and etc and hence reduces interference to following modules. Trigger Module is aimed to link up Detection Module and Tracking Module. By pre-installing virtual region, it catches the local peak of the average grey degree in a certain region according to the catachrestic of the change of grey degree in this region and locks the targets passing by, achieving auto capture of moving vehicle targets. Tracking Module is designed based on Mean Shift algorithm. It uses Kalman filtering to forecast the initial point of iteration. This method not only reduces computation, but also achieves dynamic regulation of the size of tracking window by improving the algorithm itself to adjust the size of the window of kernel function, ensuring accurate tracking in urban traffic system.This thesis stresses the research on improving detection and tracking algorithm to vehicles’ foreground. At the same time, it puts forward a trigger algorithm based on changes of gray degree in virtual region. At last, it achieves the whole detection and tracking system of vehicles based on OpenCV. Repeated researches demonstrate that this method links up the detection and tracking modules efficiently and accomplishes multi-target Auto-trigger-tracking, proving the practicality of the method and system in this thesis.