Research On Optical Flow Estimation Technology Based On Non-rigid Dense Correspondence

Optical flow is the projection of the instantaneous velocity of a pixel on a two-dimensional imaging plane of a moving object or scene surface.Not only contains the motion information of moving objects and scenes in the image,but also contains rich three-dimensional structure information.Therefore,the image sequence optical flow computing technology is one of the important research contents in the fields of computer vision,image processing and pattern recognition.It is widely used in robots,drone navigation and obstacle avoidance,medical image analysis,weather analysis and many other fields.Since the 21 st century,with the application of various effective optical flow models and novel optimization algorithms in optical flow estimation,the accuracy and robustness of optical flow calculation have greatly improved.However,when there are difficult motion forms such as motion occlusion,large-displacement motion,and non-rigid motion in the image sequence,the existing optical flow estimation model still cannot meet the challenge.In order to solve the problem of precision and robustness of difficult-to-move optical flow estimation for non-rigid and large-displacement motions,this paper mainly focuses on non-rigid dense correspondence large-displacement optical flow estimation techniques.By designing background noise and non-rigidity in image sequences.The non-rigid,densely matched optical flow model with better robustness,such as motion challenges,improves the accuracy and robustness of optical flow estimation under non-rigid and large-displacement motions.The main work of this article is summarized as follows:1.The research background and research significance of image sequence optical flow calculation technology are summarized.The development and research status of optical flow calculation technology and the main research contents of this paper are described in detail.The chapter arrangement of this paper is also introduced.2.Propose a nearest-neighbor field calculation method based on non-rigid dense correspondence of image sequences.Firstly,the nearest neighbor search is used to determine the corresponding relationship between adjacent frames;then the non-conformance area is eliminated by judging the consistency of the original corresponding local area.3.Aiming at the problem of accuracy and robustness of optical flow calculationin difficult motion modes such as large displacement motion and non-rigid motion,a non-rigid dense matching non-local TV-L1 optical flow calculation model is proposed.Firstly,the main motion pattern of the image sequence is obtained from the non-rigid nearest neighbor field in this paper and used as a clue to the motion target segmentation;then the moving target segmentation result is fused into the initial motion field of the optical flow calculation;finally,the initial motion field is calculated.As a prior knowledge of the non-local TV-L1 optical flow calculation model,the motion field is iteratively refined to obtain the final optical flow estimation result.4.Using the test image sets provided by the MPI Sintel and KITTI databases,the optical flow calculation method described in this paper and the representative optical flow calculation methods such as Classic+NL,LDOF,NNF,and Flow-Net2.0 were comprehensively compared and analyzed.Experimental results show that the proposed method can effectively improve the accuracy and robustness of optical flow estimation,especially when the image sequence contains non-rigid motion and motion background blur.In addition,this article selects different types of medical image sequences to test the practicability of this method.The experimental results show that this method has better optical flow estimation effect for CT,MRI and other medical images,indicating that this method has a wider range of applications Range and application prospects.