Moving Target Detection And Tracking Algorithm Based On Video Sequence

Moving target detection and tracking algorithm is a core issue in the field of computer vision, the underlying technology of intelligent video surveillance system. It combines the research results of pattern recognition, image processing, artificial intelligence and other fields, and has been widely applied in various fields,such as security monitoring intelligent weapons, video conference.Therefore, the moving target detection and tracking algorithm has an extremely important theoretical significance and pragmatic value.In this paper, according to the requirements of the project, we focused on the detection and tracking algorithm research for moving target in video sequence, and proposed some improved methods which also be verified by programmingIn moving target detection, this paper begins with the analysis of target detection methods, including optical flow method, frame difference method, background subtraction method and the method based on statistical learning, and point out their advantages, disadvantages and the main scope of application. Meanwhile, introduce the now popular Gaussian background modeling and theory of evidence-based information fusion background modeling algorithm principle. For different algorithms have advantage and disadvantage, this paper proposes a mixed algorithm in video sequence based on the motion target detection. Combining the median filtering background modeling and the improved Temporal Difference method (MFTD) to detect the target, which also use the self-adaptive threshold segmentation method to optimize moving object extraction. At the same time, we introduce the Gaussian filter and morphological filter to eliminate noise and improve the effect of moving region extraction.In moving target tracking, this paper first introduces the classification of tracking algorithm, including target tracking algorithm based on3D model, region matching and feature matching. Subsequently, describe the classic Mean-Shift tracking algorithm in mathematical theory level. For the deficiencies of traditional Mean-shift algorithm, this paper proposed an improved tracking algorithm based on Mean-Shift. First, a novel weighting method is proposed to avoid the shortcoming of the Bhattacharyya similarity. In order to overcome the tracking problem of fast object moving and the fixed size tracking windows could not track an object effectively when object size is changing. The algorithm also combines Mean-Shift and Kalman adaptive filter for prediction, in which adds kernel bandwidth adaptive update to enhance the robustness.Finally, the proposed algorithm is implemented in Matlab2009b, and analysis the experiment results. At the same time, summary this paper work done, and pointed out the existing problems and further research ideas.