Video-based Moving Targets Detecting And Tracking Algorithm Research

Computer intelligent video surveillance system (CIVSS) is a front topic in Computer Vision domain. It spans many subjects including computer science,machine vision,image engineering,pattern analysis,artificial intelligence,etc. CIVSS can automatically analyze the sequence of images by the methods of computer vision and video analysis. The system can detect,locate,recognize and track objects in a moving environment in real-time. Furthermore,it can also analyze and judge the movement of objects. This thesis focuses on the background modeling and moving object detection,as well as moving object tracking.Background modeling is an important issue in accurate detection of moving objects. Existing work in the area has mostly addressed scenes that consist of static structures. In this paper, we present a novel non-parametric foreground-background model which explores the complex temporal and spatial dependencies in nonstationary scenes. The model adapts to scenes which contain small motions such as tree branches and water ripple, even shadow. The Model uses GMM(Gaussian Mixture Model) to compute the probability of foreground pixel with color information. It also uses LBP(Local Binary Pattern) texture model to compute the probability of foreground pixel with texture information. At last, it uses data fusion algorithm named D-S evidence theory to do a information fusion in the decision level. Extensive experiments with nonstationary scenes demonstrate the utility and performance of the proposed approach.At the aspect of object tracking, considering the drawbacks of the current Mean Shift tracking algorithm, we present a new algorithm combining the Kalman filter with Mean Shift. And the experiment results show that the moving objects can be tracked immediately and constantly as well as the occlusion problem is well solved by using this algorithm.