Research On Reassembling Method Of Fragment Point Cloud

The massive point clouds have been obtained by convenient 3D acquisition equipment,which provides support for complicated 3D geometric calculation,making data-driven point clouds structured and semantic expression become a research hotspot in the 3D vision field.The massive point cloud data is disordered,lacking spatial geometric topology descriptions and efficient processing methods,which restrict the development of related application fields.Most of the fragments come from damaged cultural relics such as terracotta warriors,pottery,ceramics and medical bone fragments.These fragments are numerous,irregular shape structures,complex adjacency relationships,and uneven surfaces.Therefore,the geometric topology representation and feature extraction of the fragment point cloud is a more complex and challenging problem in point cloud data processing.This paper takes the fragment point cloud data as the research object,and the main research contents and innovations are as follows:(1)To address the problems of difficulty in determining the voxel granularity and low search efficiency of the existing point cloud octree representation,a topology representation method for the fragment point cloud is proposed using the 3D hash chain structure of octree(3DSL?Hash?Octree).Firstly,the leaf nodes are encoded and sorted by the breadth-first search strategy based on the octree representation of the fragment point cloud.Secondly,a fast index with the hash chain structure is designed by analyzing the spatial distribution of the leaf nodes,which relates the hash function and chain structure to the spatial topology and adjacency of the point cloud.Finally,an efficient neighborhood search method is proposed.The experimental results show that this method effectively avoids the situation that the points with similar Euclidean distance in space are mistaken for the adjacent points on the surface,enhances the detail representation ability of the fragment point clouds,and improves the search efficiency.(2)According to unevenness of the fragment surfaces,a feature extraction method based on optimal neighborhood is proposed to solve the problems of redundant feature information and complicated computation for the existing multi-scale neighborhood feature extraction methods.Firstly,the relationship between the local surface variation of fragment point cloud and the neighborhood radius is analyzed,and the neighborhood where the surface variation tends to be stable is regarded as the optimal neighborhood.Secondly,a calculation formula of projection distance is defined to describe the degree of concave-convex variation for the fragment,so that the points with large variation in concave-convex can be regarded as feature points,and two types of feature points with different projection distances are extracted.Finally,two types of feature points are clustered and refused,and the Laplacian smoothing method is used for iterative refinement.An improved polyline growth method is proposed to orderly connect the feature points into continuous and complete feature lines.The experimental results show that this method can robustly and accurately identify the feature points on the surface of the fragment point cloud,and generate the feature lines that can accurately express the shape.(3)For a large number of fragments with complex shape,the scattered feature points and the uneven feature lines,it is difficult to find the rule of feature distribution.Therefore,a coarse matching method for multiple fragment point clouds is proposed.Firstly,the thickness histogram is constructed by estimating the thickness feature of the fragment surface using the feature line of the point cloud,and the chi-square similarity of the histogram is computed to find potential matching between fragment surfaces,and the fragments with potential matching relationship are divided into a group.Secondly,in the same group of fragment point clouds,the feature vector that fuses the seven local features of the point clouds is calculated,and the covariance matrix descriptor is constructed to describe the local features of the point cloud by using the statistical properties of the covariance matrix.Finally,the similarity of descriptors is measured by using the log-Euclidean norm to achieve coarse matching of fragment point clouds.The experimental results show that this method can find the potential matching for each surface of the fragment,and the average matching rate of coarse matching of different fragments reaches 90%.The anti-noise performance of the constructed descriptor is better than that of PFPH,Spin image.(4)On the basis of the research on topology representation and feature extraction of point cloud,the research on fragment reassembling is carried out,and the automatic reassembling calculation framework for multiple fragment point cloud is designed.In the multiple fragment automatic reassembling method,firstly,the fragment are grouped according to the shape features reflected by the feature lines of point cloud.Secondly,the coarse matching is carried out by using the multi-feature fusion covariance matrix descriptor;finally,a feature refinement method with the improved RANSAC algorithm is proposed to obtain a quasi-optimal matching set,and the double-threshold method is used to finely optimize the quasi-optimal matching set,so as to realize the accurate reassembling of the fragment point cloud.A multi-fragment automatic reassembling calculation framework is constructed based on the research,and the reassembling processing is performed on the publicly available fragment datasets from the University of Technology Vienna.The reassembling results shows that the proposed multiple fragment automatic reassembling method realizes the precise reassembling,and has obvious advantages in matching accuracy and time efficiency,which verifies the feasibility and rationality of the calculation framework.