Research On Rigid Registration Algorithms Of Three-dimensional Point Clouds And Its Application In Substrate

In recent years,with the development of data acquisition technology and data processing technology,three-dimensional measurement based on vision technology has attracted more and more attention in the fields of industrial manufacturing,cultural relics protection and so on.Three-dimensional measurement involves the processing of point clouds and the acquisition of complete models.In reality,it is impossible to obtain a complete model of three-dimensional objects at one time through visual technology,so registration technology emerges as the times require.With the help of the overlapping information between the viewpoint information and the viewpoint cloud,the coordinate transformation of the relevant viewpoint point cloud is carried out,and they are unified into the same coordinate system,so that the complete threedimensional model of the object can be obtained.Focusing on the research of three-dimensional point cloud registration technology and its application in three-dimensional measurement of substrate,this paper mainly carries out the following aspects:Firstly,based on the improved process,the key technologies in each stage of point cloud registration are innovatively classified,summarized and discussed,and the problems and challenges in the field of point cloud registration are put forward in order to provide reference,inspiration or guidance for related engineering applications and academic research.The process includes:(1)In noise reduction phase,the noise is divided into three types,and the corresponding filtering methods are introduced.Finally,the general filtering strategies are summarized.(2)In the data simplification phase,the data simplification method is divided into detection method and descriptor method.The detection method is further subdivided into LRF method and non-LRF method.The descriptor method is subdivided into feature signature method,spatial distribution histogram method and geometric distribution histogram method.On the basis of this classification,several algorithms with excellent performance or distinctive characteristics are introduced.In the part of summary and discussion,we summarize the relevant performance research or comparative literature,and draw some guiding conclusions.(3)In alignment phase,alignment methods are divided into global optimal alignment method and local optimal alignment method.The rationality and feasibility of substituting global optimal alignment and local optimal alignment for coarse alignment and fine alignment in traditional process are discussed.Then the global alignment method is further divided into six categories: RANSAC,Graph,Bn B,Heuristics,Frequency and Field.The local alignment method is divided into GMM method and ICP method.On the basis of this classification,the basic principles,implementation frameworks,advantages and disadvantages of various methods are summarized,and several noteworthy algorithms are introduced accordingly.In the summary and discussion section,the types of alignment methods are summarized,and the applicability of some methods is introduced.Secondly,a three-dimensional point cloud rigid registration algorithm based on plane normal rotation and translation decoupling is proposed.The algorithm mainly includes two steps: Plane extraction and its normal calculation and processing,and optimal registration solution search.(1)Plane extraction and its normal calculation and processing,i.e.the preprocessing steps of point clouds,a set of algorithms for noise reduction,plane extraction,normal calculation and processing of point clouds on substrates are designed.(2)The search of optimal registration solution is the alignment step of point cloud.This step decouples the rotation translation: firstly,the feasible rotation solution space is determined by the normal direction of the point cloud plane,and then the solution space is compressed based on the composition law of the solution space.Finally,the optimal translation solution is searched in the rotation solution space by using the RANSAC method,the rotation solution that determines the optimal translation solution is the optimal rotation solution,and the two combine to obtain the optimal registration solution.In addition,the feasibility and validity of the corresponding method are verified by experiments.Thirdly,In the application of three-dimensional measurement of the substrate,two schemes of three-dimensional measurement based on single point cloud and error detection based on registration are designed and implemented respectively with the algorithm in this paper.In addition,the advantages,disadvantages and applicability of the two schemes are discussed.In a word,this paper investigates and summarizes the key technologies in each phase of the registration algorithm,and proposes a global optimal registration algorithm for at least two detectable non-parallel planes,such as substrates.This method has strong robustness to angle,density and other changes.In related experiments,the algorithm shows sufficient stability and achieves better results on the whole.Better results have also been obtained in the registration experiments of dealing with noise point clouds and real point clouds.