On The Occlusion-detection Based Study Of Variational Optical Flow Computation Technique With Non-local Constraint

Optical flow is a two dimension projection in image plane of moving object or moving scene in real three dimension world. It not just contains three dimension information of moving object or scene, but also contains a plenty of structure information. Optical flow computing technology in image sequence is a hot topic in the domain of image processing, pattern recognition, computer vision and so on. The research results of optical flow are widely used in robot vision, unmanned aerial vehicles navigation, vehicle auxiliary navigation, medical image analysis and so on.These years, optical flow computing technology have gain great improvement in calculation accuracy and efficiency. And it also has a lot of significant achievements and excellent algorithm. But the robustness and reliability of the optical flow computation model are still a major problem which is urgently to be solved when there are brightness saltation and discontinuous phenomenon which is caused by motion block out or large displacement motion or motion blur in image sequence. Due to the problems mention above, firstly, from energy function of optical flow computation, this article analyzed data item, smooth item and various improvement strategies of optical flow computation. Then this article found a robust model and optimization strategy of optical flow computation in difficult movement patterns. The main contribution of this article including following:1. During the computation of optical flow with the traditional method, usually it could come to a situation of trapping in local optimal solution, and this could lead to spill point of optical flow at the moving edge of a image. This article optimized the optical flow computation energy function of different improvement strategy and adds the local constraints to the basic optical flow. Then a non-square global variation optical flow computation model has been proposed which is based on non-partial constraint. Experiments show that this algorithm improved accuracy of optical flow computation at the area of moving edge in image.2. According to the influence of robustness in optical flow computation at thearea of blocking out in image, using the Delaunay triangular mesh to divide the image.According to local luminance change of triangular mesh in image sequence. Then a block out detection method for image sequence, which is based on Delaunay triangular mesh, has been proposed in this article. This is going to provide reference information for optical flow computation at moving block out in image.3. As the influence of the robustness and reliability during the optical flow computation model caused by the motion occlusion, large displacement motion and motion blur. This article combining with the motion occlusion result of image sequence based on Delaunay triangular mesh, proposed a new optical flow computation strategy based on pyramid hierarchical refinement and adaptive median filter. According to the image motion occlusion and fuzzy region to determine the weight parameters of median filter to overcome the traditional method in motion occlusion and the disadvantage of poor robustness of regional fuzzy, during the calculation of each layer image optical flow.Finally, this paper uses the Middlebury database test image set and the Sintel MPI database test image set to verify the validity and accuracy of the proposed optical flow calculation method. At the same time, this article comprehensively compares typical HS, the tv-l1, Classic++ and Classic+NL optical flow calculation methods and the proposed method in calculation precision and robustness. Comparative test results show that this method compared with other methods has higher accuracy and better robustness.