Research On 3D Point Cloud Reconstruction Based On Repeat Properties Of Building

With the development of science and technology,the new technological products based on 3D reconstruction technology,such as VR games,AR red envelopes,3D maps and so on,are changing our life.The main scenes of these applications are buildings,streets and other models,therefore how to obtain more complete 3D models is vital for these products.So this paper presents more targeted approaches to achieve complete data collecting,data processing and complete 3D reconstruction.It is common that there are trees and other objects around building,and trees will interrupt the light beam from 3D scanner leading to that 3D scanner can not collect complete point cloud of building.By the repetitive structures which are ubiquitous for building,this paper studies 3D point cloud reconstruction of building based on repeat properties of building.The study includes method for efficient point cloud registration,pretreatment of point cloud,extraction of structure of building's plane,complete generation and reconstruction based on repetitive structures.As far as large building,the commonly used method for the acquisition of point cloud is to scan many times using 3D scanner and there is enough overlap between the adjacent scans and it takes too much time for later processing.This paper presents a method for the acquisition of point cloud which combines 3D scanner and total station.The method first builds a geodetic coordinate system based on total station,and then determines the location of 3D scanner in geodetic coordinate system through the prisms which are fixed on the 3D scanner,further obtains the coordinates of geodetic coordinate system of point cloud collected by 3D scanner in different positions and achieves efficient collection and registration of point cloud.Aimming at the problem that there are grass,trees and other spurious data in the point cloud collected by 3D scanner,this paper presents an algorithm for point cloud processing,which combines projection and cluster segmentation.The algorithm presented by this paper removes grass and other low objects from the point cloud by projection,and then removes trees and other tall objects from the point cloud by euclidean cluster segmentation.In order to make up for the missing of point cloud because of occlusion of building,this paper prensents a method for complete 3D reconstruction of building.The method first uses region growing algorithm to extract the building's planes,and then obtains repeat properties of the repetitive structures of the plane by SIFT descriptor.According to the repeat properties of repetitive structures,this method divides the plane into a number of repeat units and gets a complete unit and then copies the complete unit to achieve complete 3D reconstruction of the plane.All planes are processed by this method,and then the complete 3D reconstruction of building is realized.In this paper,the feasibility of this method is verified through the experiment of the Xinyuan building of Xidian University.