Research On 3D Point Cloud Registration Based On Fusion Of Feature Description And Clustering

With the continuous upgrade of lidar and scanning equipment,3D point cloud processing technology is also booming.3D point cloud processing technology plays an irreplaceable role in the fields of 3D reconstruction,reverse engineering,medical image processing,unmanned driving and virtual reality.Since the point cloud acquisition device cannot obtain all the point cloud information of the collected object from a single Angle at one time,it needs to register the point cloud information obtained from different angles into the same coordinate system.This process is called point cloud registration.3D point cloud registration is the most basic and important part in 3D image processing.In this dissertation,the point cloud registration technology is studied from two aspects: the method based on feature matching and the method of fusion feature and clustering.The main contents of this dissertation are as follows:(1)Research on 3D point cloud rough registration method based on feature matching.Aiming at the problems of low registration accuracy and low registration efficiency without initial value,this dissertation proposes a 3D point cloud rough registration method based on feature matching.Firstly,the change of the local normal vector of the point cloud is used to describe its features,and the feature retention weight is added to screen outstanding feature information,so as to improve the registration efficiency.Then,the feature histogram is established according to the retained feature information,and the initial matching point pair is obtained by comparing the feature histogram information.Finally,the rigid invariant constraint is combined with the random sampling consistency algorithm to screen the correct matching point pairs,and then the rotation matrix and the translation vector are obtained by using the fourelement method.Experimental results show that the proposed algorithm has higher accuracy and efficiency than other coarse registration algorithms,and provides a good initial value for the subsequent point cloud precise registration work.(2)Research on 3D point cloud registration method combining geometric description and spectral clustering.Aiming at the problem that point cloud registration is sensitive to the initial value of iteration and has low ability to deal with sparse,noisy and locally missing point clouds,this study proposed a three-dimensional point cloud registration method combining geometric description and spectral clustering.Firstly,the local geometric features of the point cloud to be matched were calculated,and the adjacency matrix was constructed by Gaussian kernel and full connection method according to the local features,so as to retain the feature information of the point cloud to the maximum extent.Secondly,spectral clustering is used to divide the point cloud into point cloud information blocks with specific structural characteristics to reduce the data scale of the point cloud.Then,the feature blocks are used to restore the rigid transformation to achieve efficient and accurate rough registration of point clouds.Finally,the coarse registration result is taken as the initial value of the fine registration,and the iterative nearest point algorithm is used to achieve the fine registration of the point cloud.Experimental results show that compared with other registration methods,the proposed method can improve the registration accuracy and efficiency,and has good registration performance.