The Research Of Motion Detection, Localization And Tracking Techniques In Video Sequences

Motion detection, localization and tracking in video sequences is an important re-search topic in computer vision, which is widely utilized in many areas, such as visualsurveillance, digital entertainment, etc. The key techniques motion detection, objectlocalization, object tracking are analysed in detail in this paper, including related work-s, the proposed methods, and elaborate experiments to demonstrate the effectiveness of the proposed approaches. Some novel results are shown in these three techniques.More specifically, in terms of motion detection, a multi-layer self-organizing map basedbackground modeling method is proposed, which can extract the moving region fromcomplex background precisely. Moreover, a parallel implementation is given to guar-antee its realtime performance; in terms of object localization, a sifted probabilisticoccupancy map based object localization method and a visual hull reconstruction basedmethod called voxel projected map are proposed separately. The former one can calcu-late the probability of people’s occupancy more accurate and faster by introducing thesifting mechanism. The latter one can reconstruct the 3-dimensional contour of movingpeople in realtime by the parallel implementation and locate people more accurate; interms of object tracking, this paper implements a particle filter tracker and a k-shortestpath algorithm (KSP) based multiple zebra fish offline tracking system. The former onecan track people in laboratory scenarios by combining the particle filter technique anddesigned heuristic rules. The latter one can track multiple no-cross zebra fish automat-ically by defining a global objective function and using the KSP solver. At last, as anapplication example of those techniques, in the whole framework of the crane projectKey technology and system integration for safe and high-efficiency automated overheadcranes, this paper designs and implements a motion detection and tracking system foran overhead crane, which can detect, locate and track people in the crane workspaceautomatically. To sum up, this paper gives a deep research on motion detection andtracking techniques in video sequences, and some novel results are shown. Elaborateexperiments are carefully conducted to demonstrate the effectiveness of the proposedalgorithms. Furthermore, we design a whole motion detection and tracking system foran overhead crane, which can detect, locate and track people in the crane workspacepretty well.