The Research For Visual Surveillance System Based On Fusion Of Multiple Model Data

The visual surveillance system is the complex combination system of information science, computer science, machine vision and all kinds of hardware, technology and theories. The visual surveillance system not only provide prompt, comprehensive and the reliable information for sudden event's evidence collection, analysis and research, but also help administrators transfer the resources reasonably, and works out the pertinence solution. Thus the visual surveillance system can remarkable raises the level of security protection.Starting with the overall structure of visual surveillance system, this thesis has studied the main involved theories by utilizing the technique of intelligence information processing. Firstly, the paper proposed a novel method for moving-objects detection based on fusion of background subtraction and an improved three frames differencing in view of the influence of applications in the reality, such as light changes, camera shake, backgrounds disturbing and so on. The experimental results show that the new method effectively decreases the misjudgments caused by noise, significantly improves the accuracy, and can meet the needs of real-time. Secondly, aiming at the problem of target identification in the multi-features fusion of the visual surveillance system, a improved Dempster-Shafter evidence theory are put forward on the basis of optimization theory, which have effectively solved the problem of object recognition in the condition of high evidence conflict. This algorithm is used for video surveillance system to distinguish vehicle, people and other objects. Through analyzing the simulation example, the results show that the algorithm can gives a more reasonable combination results and has a good adaptive ability. At last, the visual surveillance system based on fusion of mult-model data is designed, which integrates motion detector module, head detector module, target recognition module and parametric active contours tracker module. The main goal of using more than one module is to make up for deficiencies in the individual modules, thus achieving a better overall tracking performance than each single module could provide.Lots of experiments are also provided for theories put forward in this thesis, including moving-objects detection and adaptive background updating method, object recognition method, and at result the effectiveness and feasibility of these methods are proved.