Research On Point Cloud Hole Filling And 3D Reconstruction In Reflective Area

3D reconstruction is the process of obtaining the three-dimensional shape or surface structure of an object,which is widely used in advanced manufacturing fields such as automotive,aerospace,industrial inspection,and reverse engineering.However,due to the structural characteristics of the component itself,the reflective properties of the coating material,and other factors,there may be specular reflection during image acquisition,making it difficult to achieve complete 3D reconstruction of the component.This paper proposes an outlier point recognition method,a point cloud hole filling method,and implements complete surface reconstruction for the problem of strong reflection and incomplete 3D reconstruction of scanned components.The main research content and results of this paper are summarized as follows:1.For the problem of outlier noise defects in the scanning of strongly reflective components,this paper studies the orderliness of outlier points in reflective areas under specific viewpoints,proposes a View-Transform-Point Net outlier point recognition network,and realizes initial point cloud alignment by identifying the initial point cloud plane and calculating the rotation pose,and then implements secondary alignment of the point cloud based on the perpendicularity between the outlier plane in mixed reflection and the point cloud plane.The two alignments improve the alignment network in Point Net and thus build a point cloud outlier recognition network to identify and extract outlier points in reflective areas.Compared with Point Net,the proposed View-Transform-Point Net improves the average intersection over union(m Io U)by 2.891% and the average accuracy(m Acc)by 4.645%,while View-Transform-Point Net++ improves m Io U by 7.867% and m Acc by 2.777% compared with Point Net++.2.This paper proposes a point cloud hole-filling method based on the principle of outlier formation to address the issue of point cloud holes in scanned highly reflective objects.Firstly,the length of each outlier plane projection on the xoz plane of the reflection area is obtained in the point cloud space,and the corresponding structured light grayscale value of the same pixel position at the beginning and end of each outlier plane in the image domain is obtained.Then,based on the high Gaussian distribution characteristic of structured light intensity,a local Gaussian distribution is approximated to linear variation and is linearly correlated with the outlier plane.Finally,the distance between the end of each outlier plane and the real surface is solved to repair the depth information of outlier points.Compared with the unfilled method,the number of point cloud filling is increased by 39.4%,the number of triangular mesh faces is increased by 45.2%,the surface area is increased by 46.9%,and the chamfer distance(CD)of the point cloud is 0.4471009.Compared with existing geometric repair methods,the standard deviation is smaller,the smoothness is stronger,and it better fits the real surface characteristics of the object.3.In order to achieve complete 3D reconstruction of the repaired point cloud model,this paper first calculates the point cloud normal vectors,discretely stores the point set in function space,then solves the isosurface through the Poisson equation,and finally completes the remaining holes by adjusting the parameters,thus achieving the 3D reconstruction of the point cloud model.Compared with the Delaunay surface reconstruction method,the number of triangles in the mesh is increased by 97.9%;compared with the registration error of the original model,the average distance is increased by 5.26%,and the standard distance is increased by3.43%;compared with the unrepaired model,the calculation time is reduced by 256 msec.In terms of repair effect,this method reduces the error caused by large curvature in the boundary region,improves the smoothness,and reduces sharp edges.