Research On 3D Point Cloud Data Registration Technology

With the continuous development of computer vision,3D reconstruction technology is widely used in industrial manufacturing,cultural relics restoration,reverse engineering,virtual reality and other fields.Three-dimensional reconstruction technology is often used to compare and analyze models in various fields.However,due to the influence of scanning equipment,acquisition environment,object surface reflectivity and other factors,it is impossible to obtain all the point cloud information of the object to be measured at one time.It is necessary to scan from multiple angles and transform the results into the same coordinate system.The process of transforming point clouds from different angles into the same coordinate system is called point cloud registration.In this paper,the registration method based on the local geometric features of point cloud is studied,and the existing method is improved.The main work is as follows:Aiming at the problem that there are many wrong corresponding point pairs in the source point cloud and the target point cloud,which affects the registration accuracy,a point cloud registration algorithm based on optimized screening of feature point pairs is studied.Firstly,the principal component analysis algorithm is used to approximate the curvature.Secondly,feature points are selected and the feature histogram of two-point set is established.Thirdly,based on the results of feature description,the corresponding point sets of two-point clouds are initially selected;Then,the corresponding point sets are optimized and screened by using the one-to-one correspondence principle and the approximate equidistant principle;Finally,Rodriguez rotation formula is used to solve the rotation matrix and translation vector.The experimental results show that this algorithm can reduce the registration error compared with the traditional RANSAC registration algorithm which eliminates the wrong point pairs.Feature point extraction algorithm based on approximate curvature value will keep more similar information points as feature points of point cloud,which will affect the accuracy of registration algorithm.To solve this problem,a registration algorithm based on point cloud feature fusion is studied.Firstly,the Gaussian curvature of the input point cloud is obtained.Secondly,combining Gaussian curvature with point cloud density,the fusion eigenvalue of point cloud is obtained,and the initial feature points of point cloud are selected based on this eigenvalue.Then,the constraint condition of normal angle is added to screen the initial feature point set,so as to extract the feature contour points of point cloud.Finally,point cloud registration is completed by using the extracted feature points.The experimental results show that this algorithm can improve the registration accuracy compared with several common feature point extraction registration algorithms.