Study On Moving Object Detection And Tracking In Sequence Images

In recent years, with the continuous development of computer video processing technology, more and more researchers begin to pay attention to computer vision at home and abroad. The moving object detection and tracking as an important computer vision technology has a broad application prospect in transportation, military, public security and other field. Therefore, it is of great theoretical significance and practical value to conduct the thorough research on the computer vision. The main research work on the moving object detection and tracking algorithm carried out in this paper are listed below:Firstly, in terms of the moving object detection. This paper studies an object detection algorithm based on the turbulence model on the basis of the analysis of the classic object detection algorithm. By which we first establish the turbulence model with pixel intensity and movement information, and get an object confidence map. We then have three-term rank decomposition on the video sequence, and decompose it into the background, target and turbulence. We leverage low-rank, the Gaussian and the sparse properties of the video sequence caused by atmospheric turbulence deformation decompose it under certain constraints, and then the sparse solution of this turbulence model is positioned in the non-Gaussian movement area.Secondly, in terms of the moving object tracking. We firstly introduce classic object tracking algorithm, and then study an object tracking algorithm based on the scale and orientation adaptive on the basis of the Mean Shift algorithm. In this method we first estimate the scale of target by the zero matrix and Bhattacharyya coefficient. We then calculate the covariance matrix reflecting the scale and orientation of the target. We later estimate the real scale and orientation of the target form the candidate region.Finally, this paper studies an object tracking algorithm based on joint registration and active contour segmentation on the basis of the moving object detection and detection algorithm. In this method we firstly estimate the affine deformation of the target using the registration procedure, and then evaluate non-rigid motion of the target. We then improve the affine deformation using segmentation procedure, and then accurately calculate thereal contour of the target. We later remove tracking drift by the on-line target appearance updating.