The Research And Implementation Of Target Detection And Tracking Under Moving Background

Video-based moving target detection and tracking technology under moving background is an important research direction in computer vision. The moving target detection and continuous tracking with the moving camera platform is required in many practical applications. It could expand the tracking range and reduce the cost and limitation of building many target tracking devices with stationary cameras. Compared with the stationary camera, the background motion is joined into the video frames because of the moving camera platform and increases the complexity of moving target detection and tracking, so the research of the moving target detection and tracking under the moving background becomes more important.After studying the key technology of the moving target detection and tracking, this paper focuses on the research of moving target detection based on the background motion estimation and the tracking as following.In moving target detection, this paper proposes a method based on the KD-tree and Multiple-Constraint to match the feature corners roughly, and then uses the covariance control to get more accurate comer matching. After matching the corners, RANSAC and Least Squares are used to robust estimate the parameters of background motion under the affine parameters model. After the compensation of the background with the global motion parameters, this paper uses frame difference and background subtraction to detect the moving target, analysis the result of the two methods and accept the frame difference in the target detection to acquire the target area for target tracking finally.In moving target tracking, this paper improves the MeanShift tracking method that uses the combine of two-dimension color histogram and two-dimension structure histogram with gradient and position features as the target model. This improvement enhances the target tracking when the target has the same color distribution with the background. This paper combines the MeanShift and Kalman filter to predict the target trace and realizes the robust tracking while the part of the target is blocked.Experimental results show that the method this paper improves could detect and perform robust and continuous target tracking under moving background.