Design And Implement Of An Intelligent Visual Surveillance System

With the improvement of living standards and computer vision technology, there is a growing emphasis on emergency prevention and emergency response. Video surveillance which can automatically detect, track and identify moving targets in the scene by analyzing the video image sequence without manual intervention, is one of important means of achieving safety monitoring. Moreover, it makes the basis of carrying out much higher level processing of understanding and analyzing the behavior of objects. The key technologies in visual monitoring are moving targets detecting and tracking, which have a wide application prospect.The visual surveillance system mainly includes two parts, one is moving targets detection and extraction, the other is targets prediction and tracking. Based on studying the algorithms for moving targets detection and tracking, the main objective of this dissertation is to build an experimental tracking system. The major work of this thesis is as follows:1. Frame difference and background subtraction, which are commonly used algorithms in visual surveillance for moving targets detection, are comparatively studied. The principles and procedure of detection are given. Frame difference algorithm and Gaussian Model are compared through experiments. The results demonstrate the advantages and disadvantages of the two methods, and give the application areas of them. Gaussian Mixture Model (GMM) is adopted for background modeling in this work.2. Basic principles of the Independent Component Analysis (ICA) and related mathematical conceptions are introduced. A novel ICA based approach of moving targets detection is proposed, and under the environment of MATLAB 7.0, the simulation experiment based on the Infomax-ICA algorithm is used. It is shown that the approach of moving targets detection based on ICA is an effective detection methods. 3. In the aspect of object tracking, the CamShift algorithm and Kalman motion estimation are mainly studied in this thesis and the principle and process of the two methods are introduced in detail. This laid the theoretical foundation for the design of the system tracking module.4. An intelligent visual surveillance system based on the OpenCV is designed and completed. In this thesis, the overall structure of the system is firstly determined according to the module requirements. Then under the environment of VC++ 6.0, tracking software is implemented with OpenCV library and an intelligent visual surveillance system which adopts modularization design, is built using equipment of our laboratory. The results are given under experiments on both our own and standard video sequences. It is shown that this system can successfully realize motion detection and tracking mission, with good robustness and real-time performance.