Research On Automatic Point Cloud Registration Method Based On Local Features

Three dimensional laser scanning technology has become one of the main means of Surveying and mapping technology because it can reconstruct the scanned object completely and accurately,access the original measurement data quickly,and get close to the object prototype quickly.The registration technology of the 3D point cloud is the core link in the entire point cloud data processing process,which directly affects the effect of the 3D reconstruction of the measured object.In the context of increasing requirements for surveying and mapping results,the requirements for point cloud registration technology have also increased.Continued in-depth research on point cloud registration technology has important significance and effect on the development of surveying and mapping technology and the advancement of other related fields.On the basis of analyzing some of the advantages and disadvantages of the current registration algorithm,in view of some of the existing problems in the current point cloud registration research,the following researches are carried out on the problems faced by the current point cloud registration technology:(1)For the three-dimensional point cloud data containing color attributes,the traditional multi-scale covariance(MCOV)descriptor method can effectively identify and match three-dimensional surface features,and find the corresponding relationship through geometric constraints to complete the three-dimensional point cloud.Accurate registration.However,because the covariance matrix is multi-dimensional data,when the amount of point cloud data in the three-dimensional scene is large,the time complexity of the MCOV method to extract key points is relatively high,which results in the entire point cloud registration process is very time-consuming.This article proposes a method to extract key points based on the change of the local feature angle of the normal vector.According to the degree of change of the local normal vector of the point cloud,select the local feature with a large change as the key point.Then the multi-scale covariance descriptor is used to find the corresponding relationship,and the game theory method is used to eliminate the wrong corresponding relationship.Experimental results show that this method can quickly find the ideal key points.Compared with the MCOV method,the method proposed in this paper can effectively reduce the key point extraction time,and its registration efficiency is significantly improved.While improving the registration efficiency,the registration accuracy and sensitivity to noise of the method in this paper are consistent with the MCOV method.(2)For the point cloud without color attribute,because the color information is not added to the constraint conditions,the constraint conditions are reduced,and only using the geometric characteristics of the point cloud to establish the covariance matrix for point cloud registration is bound to affect the registration accuracy.Therefore,on the basis of covariance,a coarse registration method based on covariance eigenvalues is proposed for point clouds without color information.This method first uses voxel downsampling to streamline the point cloud data to improve the registration efficiency.Then,the covariance matrix is constructed through local neighborhood features,and the eigenvalues and eigenvectors of the matrix are calculated according to the characteristics of the covariance matrix.And perform normalization vector processing on the eigenvalues to find the first correspondence.Due to the few constraints and low accuracy of the initial correspondence,the precise correspondence is found through correspondence propagation and the final correspondence is determined by combining the random sample consensus mismatch elimination method to complete the initial registration.Through experimental comparison and analysis,this method can effectively complete the initial registration of point clouds,improve the registration accuracy,provide a good initial state for fine registration,and achieve satisfactory results;in addition,compared with other methods,the applicability of this method is more Wide,in the case of low point cloud overlap,the method in this paper can still complete the point cloud registration excellently.Figure[37]table[6]reference[79]...