Research On Variational Optical Flow Technology Of Image Sequence Based On Alternating Guided Filter

Optical flow refers to the instantaneous velocity of a pixel in a two-dimensional projection plane of a moving object or scene in three-dimensional space.It not only contains the motion information of the target object or scene,but also carries information such as the three-dimensional structure of the target object or scene.Therefore,the study of image sequence optical flow computing technology has become a hot topic in image processing,pattern recognition,computer vision and other fields,and these research results are widely used in robot vision,drone navigation,vehicle-assisted driving,traffic video surveillance,medical image analysis,virtual reality and other fields.With the optimization of various new methods and new technologies applied to optical flow models in recent years,researches on optical flow calculation technology has achieved rapid development and has been greatly improved in terms of computational accuracy and efficiency.However,when there are difficult scenes and motion forms like noise,moving edge blurring,abrupt change of light luminance,shadows,non-rigid movements,large displacements or etc.,the accuracy and robustness of the optical flow calculation will be greatly affected.For the problem of edge blur in image sequence optical flow calculation,this paper mainly focuses on the technique of image sequence optical flow calculation based on alternating guided filtering.By designing an optical flow optimization model for sequence images based on alternating guided filtering,it overcomes such problem as optical flow calculation edge blurring caused by noise,large displacement and non-rigid movements in order to improve the accuracy and robustness of the optical flow calculation.The main work of this thesis includes the following points:1.It introduced the research background and significance of image sequence optical flow technology and focused on the development and research status of optical flow computing technology,and analyzed in detail the problems and challenges faced by current optical flow technology.2.It analyzed classic optical flow calculation methods such as HS and CLG-TV and typical optical flow methods like Classic+NL and NNF proposed in recent years.It also introduced commonly used strategies and filtering methods in variational optical flow technology.3.For the problem of image and motion edge blurring caused by difficult scenes such as noise,large-displacement motion and non-rigid motion,or etc.,an optical flow calculation optimization method based on alternating guided filtering was proposed.In the process of optical flow calculation,an alternating guided filtering optimization model is added after the calculation of optical flow in eachlayer of the hierarchical optical flow framework of the pyramid,and Rolling Guidance Filter(RGF)strategy and Smooth and Iteratively Restore filter(SIR)strategy are used through alternate iterations.This method can well protect the edge information of the optical flow field and filter noise and abnormal value optical flow.It also overcomes the problem of edge blurring of the optical flow field,and improves the accuracy and robustness of the optical flow calculation.4.Evaluation criteria of average angle error(AAE)and average end-point error(AEE)as well as the standard test image sets provided by Middlebury,MPI Sintel and KITTI database are respectively applied to the contrast test of alternating guided filtering optimization for four classic and representative variational optical flow methods,namely,HS,CLG-TV,Classic+NL and NNF.The experimental result shows that when the image sequence includes noise,abrupt change of light luminance,shadows,non-rigid motion and large displacement motion,the optical flow calculation optimization model based on the alternating guided filtering proposed in this paper can effectively improve the accuracy and robustness of optical flow calculation.