Research On Point Cloud Registration In 3D Reconstruction

At present,the continuous improvement of computer hardware and software performance has greatly promoted the development of 3D reconstruction technology.In general,3D scanning equipment can only reconstruct 3D point cloud with a single perspective at a time,so in order to obtain a complete point cloud model,we must splice point clouds with different perspectives,and this splicing process is point cloud registration.Point cloud registration technology is one of the most important point cloud processing technologies in 3D reconstruction.It is through calculating the transformation matrix between two pieces of 3D point cloud,so that the two pieces of point cloud can be transformed to the same coordinate system,and the overlapping parts of point cloud can be spliced.This thesis focuses on the research of point cloud registration technology in 3D reconstruction.Firstly,it introduces the basis of point cloud registration,as well as the establishment of experimental system to collect point cloud data and preprocess it;secondly,it studies different point cloud coarse registration methods;finally,it studies the point cloud fine registration based on ICP algorithm,and realizes the continuous registration of multiple point clouds on the basis of point cloud pairwise registration.The main contents and innovations of this thesis are as follows:1.Introduce the basis of point cloud registration,mainly including the concepts related to point cloud and point cloud registration,the acquisition method of point cloud,the pretreatment of point cloud,the mathematical model of rigid body transformation and the solution of rigid body transformation matrix.2.To study the method of point cloud coarse registration,firstly,two kinds of point cloud coarse registration algorithm based on RANSAC algorithm are studied,and the registration effect and efficiency of the two methods are compared and analyzed.Aiming at the disadvantage of the low efficiency of the two methods,this thesis proposes to improve it by combining the key point extraction technology.Aiming at the problem that the two rough registration algorithms based on RANSAC idea are easy to fail in rough matching of point cloud models with high structural similarity from all angles,this thesis proposes a rough registration method using rotating platform to solve the problem.3.Point cloud fine registration is studied.This thesis is mainly based on ICP algorithm.First,the basic principle and steps of traditional ICP algorithm are introduced.Then,the whole process of ICP algorithm is divided into five stages,and different strategies of each stage are studied respectively.Then,the principle and experiment of error point pair removal and error function construction are introduced in detail.In the stage of constructing error function,a new method combining "point to point" and "point to surface" is proposed.Through the analysis and comparison of different methods in each stage,the most suitable strategy is selected,and an improved ICP algorithm flow is proposed to improve the registration accuracy.Finally,based on the point cloud registration,this thesis realizes the continuous registration of multiple point clouds,and solves the problem of error accumulation through the global optimization algorithm.