Research On Registration And Reconstruction Of 3D Point Cloud Scene

With the development of point cloud acquisition equipment,generating realistic 3D digital models based on various acquisition equipment has become one of the main means of acquiring object surface information,and has been widely used in computer vision,aerospace,3D printing,terrain mapping,virtual Reality,archeology,reverse engineering and other fields.So the point cloud registration and 3D reconstruction technology are particularly important.This paper studies and improves point cloud registration algorithm and point cloud 3D reconstruction algorithm.Firstly,aiming at the problem that there are a lot of wrong point pairs in the traditional RANSAC(Random Sample Consensus)coarse registration algorithm,which affect the registration accuracy and convergence speed of point cloud,an improved random sampling consistent registration algorithm without threshold based on the elimination of wrong point pairs is proposed.Using edge detection algorithm and distance constraint method combined with improved RANSAC algorithm without threshold to eliminate the wrong matching point pair,the final matching point pair will be obtained for point cloud rough registration.Furthermore,the coarse registration accuracy of point cloud is improved,which provides a good initial position for the next step of fine registration.Then,aiming at the problems of low registration accuracy and long running time of traditional ICP(Iterative Closure Point)algorithm,the paper improves the corresponding point pair search and error corresponding point pair elimination,and proposes an improved ICP algorithm based on KD tree corresponding point search,TrICP(Trimmed Iterative Closure Point)algorithm and normal vector included angle threshold to eliminate error point pair.The experimental results show that compared with the traditional ICP algorithm,the improved ICP algorithm improves the registration accuracy and convergence speed.Finally,an improved greedy projection triangulation algorithm is proposed to improve the reconstruction effect of greedy projection triangulation algorithm.The redundant point removal algorithm is used to remove the overlapped points and uniform the point cloud density.Then the MLS(Moving Least Square)algorithm is used to smooth the point cloud.Finally,the MLS algorithm is used to estimate the normal vector for greedy projection triangulation reconstruction,and good 3D reconstruction results are obtained.