Research On Layering RBGD Image Sequenses Scene Flow Estimation Based On FRFCM Clustering And Depth Information Optimization

The 3D scene stream is a three-dimensional motion vector formed by a spatial scene or an object moving,which not only contains three-dimensional motion parameters of the scene or the object,but also embodies rich three-dimensional structure information.By calculating the continuous frame image sequence 3D scene stream,the three-dimensional motion and structure of the scene and the object can be effectively restored.Therefore,3D scene stream computing technology is a research hotspot in the fields of image processing,computer vision and artificial intelligence.The research results are widely used in UAV obstacle avoidance,target tracking,weather prediction,driverless,intelligent traffic detection command,and intelligence.Robots and other fields.In recent years,with the marketization of consumer-grade depth cameras such as Microsoft Kinect and Intel RealSense,image depth information acquisition has become more convenient.Therefore,scene stream calculation methods based on RGBD image sequences have gradually become the focus of 3D scene stream computing technology research.Although the RGBD scene flow model has made great progress in calculation accuracy and reliability,when the scene contains large displacement motion,local complex structure and motion occlusion,it often leads to edge blur and local over-smoothing in RGBD scene stream estimation..In order to improve the accuracy and robustness of RGBD scene stream estimation,this paper mainly studies the layered RGBD image sequence scene stream computing technology based on FRFCM clustering and depth information optimization.The research work of this paper includes the following points:1.The research significance and background of scene stream computing technology are summarized.The research status of scene stream computing technology is introduced in detail,and the key problems in scene stream estimation are listed.2.This paper introduces the basic theory of scene flow calculation,briefly describes the relationship between optical flow and scene flow,introduces two classicscene flow calculation models in detail,and analyzes the optimization strategies in RGBD scene flow estimation.3.Aiming at solving the problem of edge blur and local over-smoothing in RGBD scene stream estimation under large displacement motion and local complex scene,a hierarchical RGBD scene stream computing method based on FRFCM clustering and depth information optimization was proposed.Firstly,according to the FRFCM theory,the optical image of the optical flow is clustered and segmented to obtain the initial segmentation image.Secondly,the segmentation region is obtained by using the depth information with higher confidence to perform the second segmentation of the segmentation image to obtain the final image segmentation region.The prior information is combined with the segmentation result to calculate the motion of each layer to obtain the initial scene stream.Finally,the final scene stream calculation result is obtained by using the deformation technique and the QPBO fusion algorithm to refine the layers of the motion from coarse to fine.4.In order to verify the effectiveness and advancement of the proposed method,the algorithm and the representative methods such as LDOF,Sun-RGBD and Classic++ are synthesized by using the scene stream test image sets such as Middlebury2003,Middlebury2005,MPI-Sintel and SRSF.Comparative analysis.The experimental results show that the proposed algorithm has higher accuracy and robustness than other methods,especially in the large displacement motion and local complex structure image sequences,which can better overcome the edge blur and excessive smoothing.