The Research On Detection And Tracking Algorithms Of Video Motion Objects

The detection and tracking algorithm of video motion objects is one of hotspots in the field of computer vision. It's also the key intelligent technology of video surveillance system. It concerns the research results of many fields such as pattern recognition, image processing, and artificial intelligence, and so on. It has wide application in security monitoring, intelligent weapons, video conference and video retrieval et al. So, detection and tracking algorithm has important theoretical significance and practical value.This paper is under the develop request of the horizontal scientific research project-the monitoring system of early warning and tracking on the whole senses video. This thesis is mainly done some researches on the detection and tracking algorithm of video motion objects, based on application background of the intelligent video surveillance. Some new methods are put forward in this article, which implemented and certificated the effectiveness through programming.In terms of motion objects detection, three common approaches in this field of motion object detecting are analyzed including optical flow, temporal differencing and background subtraction. Then their relative merits and main application range are pointed out. The three Frames subtraction and the background subtraction of Gaussian mixture model are researched. The three Frames subtraction is relatively simple, which has strongly adaptability to the environment. However, the motion objects detected are inaccurate. The background subtraction of Gaussian mixture model can extract the background model and update time by time in the motion object scene. But when the scene's global illumination mutate, the entire video frame will be detected as a motion object, resulting in phenomenon of false detection.In the motion objects tracking aspects, four commonly used algorithms are summarized, including tracking algorithm based on 3D model, tracking algorithm based on regional matching, tracking algorithm based on feature matching and the tracking algorithm based on active contour tracing. Focuses on introducing the Mean Shift tracking algorithm, which belongs on tracking algorithm based on feature matching. To be directed against it lacking target model updating algorithm and the overall target model updating algorithm appearing model drift problem, this paper proposes a selective model updating algorithm applied to Mean Shift motion objects tracking algorithm. The core of the algorithm is that firstly calculate the matching contribution of each component in target model on the whole similarity measures coefficient, and then to selectively update the each component of the target model according to their value. The improved algorithm can effectively and steady track the motion objects, with good tracking and robustness.Finally, the detection and tracking algorithms of video motion objects are programmed to achieve on the software platform of Visual C++6.0, using OpenCV library functions. Then the results of the program realization are analyzed. The motion objects detection algorithms include inter-frame subtraction method, three frames subtraction method and background subtraction method of Gaussian mixture model; motion objects tracking algorithms include traditional Mean Shift tracking algorithm, Mean Shift tracking algorithm with the overall model updating and the selective model updating proposed in this paper.