Research On Multiple Moving Objects Real-time Detecting And Tracking In Complex Scene

In recent years, with the popularity of video surveillance systems, especially after the "9.11" incident and the bomb attacks in London subway, the investment in safety monitoring is in sharp rise. Therefore, the research on moving target detection and tracking technology, become a very important and valuable topics. Through intelligent monitoring method, not only save the labor cost, but also protect the social security.The question on studying real-time moving object detection and tracking problem in the complex scenes is discussed in this dissertation. Firstly, it describes the background ,the research status , as well as the classification of the target detection and tracking, and then it leads to the focus of this thesis: the research on the moving objects detection and tracking based on the background subtraction of a single static camera, which includes three basic steps :background extraction, target detection and target tracking. They are the skeletons in it.In the respect of background extraction, two widely used methods—Gaussian mixture background and Kalman background—has been studied in detail and analyzed experimentally. Then the concept of the convergent frame has been put forward according to the characteristic of Gaussian mixture background model—fast in convergence while complicated in computing, and that of Kalman filter—slow in convergence while simple in computing. After determining convergent frame, Gaussian mixture model is used to initialize the background before it, and Kalman filter is used to update the background after it, therefore increasing the speed of extracting the background and enhancing the real-time processing performance of this system.In the respect of target detection, four basic methods have been summarized firstly: background subtraction, frame difference method, optical flow and feature matching method, and experimental analysis of the applicability of these methods have been given at the same time. Then several mathematical theories for threshold selection have been studied in depth: fixed threshold, Otsu threshold and Renyi entropy threshold, and the introduction of genetic algorithms and ant algorithm to optimize the search on the threshold, and also by experimental contrast, a conclusion has been drawn—in complex scenes threshold on background subtraction image can improve the detection quality. Finally, focus on the post-processing method to eliminate the shadow method, and eliminate shadow to the prospects image after threshold by experiment and reach the final test results.In the respect of target tracking, the principles of particle filter have been studied firstly, and the applications in target tracking have been introduced in this section. Secondly a number of measures to optimize the process of sampling and re-sampling are proposed based on analyzing the process of the particle filter. Then after the analysis of occlusion relations a method to determine block and a method to shape-based color matching auxiliary particle filter tracking have been proposed to solve the tracking block. Finally, the experimental analysis of the effectiveness has been given, and some simple data analysis based on the tracking the target trajectory has been given: calibration, calculation of quasi-world coordinates and calculation of speed and acceleration.In summary, the results above not only solve the moving target detection and tracking technology in several key issues, but also enriched the theory methods in this respect and implementation techniques.