Research On Registration Algorithm Of Point Cloud Data In 3D Reconstruction

Along with the development of co mputer technology and 3D scanning technology, the 3D model reconstruction technique is widely applied to computer aided design, virtual reality, surveying, medicine and other different fie lds. In the fie ld of 3D point cloud reconstruction, point cloud registration technology is the core work of the 3D point cloud reconstruction process. When acquiring the three-dimensional point cloud data, due by measuring equipment and of the object to be measured and impact of various factors, cannot get a one-time measured object a full range of geometric information, so it is necessary to get 3D data of the object to be measured from different point of view. The 3D data will be obtained from different angles into a coordinate system, so as to form a complete point cloud data model, namely point cloud registration. The purpose of point cloud registration is to find an optimal solution to the point cloud data from different points of view, and to maximize the local data alignment and integration into a complete 3D point cloud model.In this paper, the 3D point cloud data registration algorithm is deeply studied, and a method of init ial registration and accurate registration based on geometric feature points is proposed:First, the three-dimens ional point cloud init ial regis tration algorithm, the computation time is time-consuming, the registration effect is not ideal, can not be applied to large-scale point cloud problem. In this paper, we propose a RAN SAC registration algorithm based on geometric features. The algorithm can extract the characteristic points of the specific area and simplify the point cloud by using the weighted processing of the ridge and valley points of the 3D point cloud, which can simplify the order of the point cloud, and improve the effic iency of the search. Registration method based on RANSAC thought in the init ial registration process, by random three sampling points of the triangle, the length restrictions, a substantial increase in the number of iterations and the approximate congruent triangles, improve the accuracy of the initial registration.Then for the tradit ional iterative closest point fine registration algorithm, based on the curvature as geometric feature extraction feature points, susceptible to noise interference, affect the registration accuracy problem. In this paper, we propose a method of selecting points to face as the corresponding point, and increase the application range of the algorithm, and calculate the sum of the Euclidean distance between the sampling points and the three neighboring points. While adding threshold limits its scope to exclude noise on corresponding point on the impact of the selection and increase the convergence speed of the algorithm, improve the registration accuracy.Finally, the registration results are demonstrated by experiments, and the effectiveness of the algorithm is verified by experiments.