Research Of Moving Object Detection And Tracking Algorithm In Video Sequences

Intelligent video surveillance system has gained significant attentionby means of its wide application prospect in computer vision. As the mostimportant foundation of the intelligent video surveillance system, detectionand tracking of moving object has always been a hot study in recent years.However, the diversity of the target and the complexity of the environmentmake the detecting and tracking become too difficulty. This paper focuseson some specific problem in the process of detecting and tracking themoving target. The major works of this paper are summarized as follows:In the detection of the target, detecting algorithms based oninter-frame difference and background subtraction have been in-depthstudied. The advantages and disadvantages and the applications of thealgorithm have been analyzed. According to the characteristics of these twoalgorithms, an improved algorithm combined the three frames differenceand the background subtraction of Gaussian mixture model is proposed inthis paper. Meanwhile, the updating method of the background model isalso improved. The new method not only can improve the detection resultssuch as big and slow moving object but also improve the convergencespeed of the background model. So, it can extract the moving objectcompletely in real time, and make a good foundation for the followingtracking task.In the tracking of the target, Camshift tracking algorithm, which isbased on color feature matching, and estimation method of the target statebased on Kalman filter have been in-depth studied. The advantages anddisadvantages of traditional Camshift algorithm have been analyzed indetail. An improved tracking algorithm based on Camshift and Kalmanfilter is proposed to deal with the problems in traditional Camshiftalgorithm, such as increasing possibility of tracking failure when occlusionor similar color interference appears. Local difference and improvedKalman updating strategy were introduced in new method. When target andbackground have a large area of similar color, local difference is used toeliminate the effect of static background pixels for tracking. When severe occlusion appears, the Kalman filter is updated by the Kalman predictivevalue instead of the value calculated by the Camshift algorithm. Theexperiment results demonstrate that the improved algorithm can track thetarget effectively and has better robustness to color interference and severeocclusion.