Research On Point Cloud Mosaic Algorithm In 3D Reconstruction

With the rapid development of three-dimensional scanning equipment and computer-aided design technology,all walks of life are faced with the development of information,intelligent and modern needs,making the three-dimensional reconstruction technology has been widely used.This paper mainly studies the point cloud stitching technology in 3D reconstruction,and studies how to improve the efficiency of point cloud reduction and point cloud registration by using scattered cloud as a research object.A point cloud reduction algorithm based on normal vector angle and an improved ICP algorithm for quaternion are proposed.The main work of thesis is as follows:Firstly,the study of point cloud data collection and processing technology.In this paper,the different point cloud data of point cloud data acquisition methods are classified in detail,and the construction method of topological relations between point cloud data is studied,including octree method,KD-tree method and rasterization method.In the end,the advantages and disadvantages of each method are analyzed.Secondly,a point cloud reduction algorithm based on normal vector angle is proposed.In this paper,we use the normal vector feature of point cloud data to calculate the key value of each point and its k nearest neighbor points.We introduce the concept of key degree and calculate the key value of each point.By comparing with the preset threshold,complete the rough reduction,and reduct the point cloud data for different proportions of the second streamlined,and finally form the theoretical and experimental point of view to verify the correctness and efficiency of the algorithm.Thirdly,through the analysis of the principle of traditional ICP algorithm,an improved ICP algorithm based on unit quaternion is proposed.Using KD-tree to search the nearest neighbor neighborhood,the search speed of the corresponding point is accelerated.Based on the curvature,the concept of matching degree is proposed to measure the degree of matching between point pairs,and the wrong point pairs are removed by setting the threshold.The rotation matrix and the translation matrix are solved by the unit quaternion method,and the distance between the nearest two points is used as the condition of the iteration to complete the final registration.As a result of the KD-tree search for the nearest point,which to some extent increased the point of the matching speed,shortening the overall time-consuming cloud registration.At the same time,the error matrix is removed according to the matching degree,and the transformation matrix is solved by using the unit quaternion method several times,which greatly improves the accuracy of the point cloud registration.Finally,the performance of the algorithm is verified by experiment.