Research On Key Technology Of Target Detection And Tracking In Intelligent Surveillance System

In recent years, intelligent video surveillance technology has been widely used in various fields. Moving target detection and tracking is the key technology of intelligent video surveillance system, and has been the hot and difficult problems of academic research. Because of background changing, lighting conditions, occlusion, shadow, and the complexity of objects modeling in real scenes, target detection and tracking are still facing many problems both in theoretical study and practical applications.Based on the comprehensive analysis of existing target detection and tracking algorithm, this thesis extends the research mainly on two aspects:Target detection and tracking in dynamic background, and multi-target tracking problem. The main work of this thesis includes the following four aspects:(1) In this thesis, key technologies of intelligent video surveillance are summarized. The principle, advantages and disadvantages of currently used target detection algorithm (optical flow method, two-frame subtraction method, third-frame subtraction method, simple background subtraction method and the Gaussian Modeling method) are compared and analyzed; The classification and principles of currently used tracking methods are studied, the typical tracking algorithms: Particle Filter, Meanshift and Kalman Filter are explored, and current difficult problems of target tracking are summarized.(2) Aiming at the objects detecting problem in dynamic background, a detection algorithm based on the estimation of background motion parameters is presented to detect motion objects in dynamic background in this paper, and the validity of the algorithm is verified by experiment.(3) Aiming at the unsatisfied tracking robustness of Camshift in complex and dynamic background, an improved Camshift algorithm, which can be apply to dynamic background, is proposed.(4) In order to improve the demerit of Camshift that it only can be used in single object scenes, and to track the static object, out-scene objects, re-moving objects and re-enter objects in real applications, an improved algorithm based on the combination of motion trajectory and Camshift, is presented.