Study On Non-Rigid Registration Algorithms Of 3D Point Clouds Based On Improved ICP Algorithm

Three-dimensional(3D)point cloud registration is to find the optimal spatial transformation relationship between two or more 3D point clouds with the purpose of achieving a good match in space.It is one of the key technologies in several domains,such as computer vision,pattern recognition and intelligent robot,etc.And it has great theoretical significance and practical value.3D point cloud registration can be mainly divided into two categories: rigid registration and non-rigid registration,the study on the non-rigid registration algorithm of point cloud with noise,outliers and missing data is still a hot and difficult problem in the current point cloud registration field.In terms of non-rigid registration of 3D point cloud with noise,outliers and missing data,this dissertation presents similarity registration algorithm and affine registration algorithm of 3D point cloud based on the improved iterative closest point(ICP)after deeply studying the traditional ICP algorithm.The main works include two aspects:(1)Aiming at the problem of similarity registration of 3D point cloud data with heavy noise and many outliers,a similarity registration algorithm of 3D point cloud based on improved ICP is proposed.Considering that the pseudo-Huber loss function can suppress both noise and outliers,an optimization model of similarity registration of 3D point cloud is established by introducing the pseudo-Huber loss function,so as to improve the accuracy of similarity registration of 3D point cloud.On this basis,the initial scale factor is solved by using the scale consistency criterion.And the similarity registration problem can be transformed into the rigid registration problem.Finally,according to the value interval of scale factor,rigid transformation matrix and shift vector are iteratively solved,and the optimal scale factor and rigid transformation can be achieved.The experimental results show that,the proposed algorithm can effectively deal with the similarity registration problem of3 D point cloud including noise and outliers,and the similarity registration accuracy is high.(2)Aiming at the problem of poor robustness and accuracy of general affine registration algorithms of 3D point cloud with noise and missing data,an affine registration algorithm of 3D point cloud based on improved ICP is proposed.On the basis of the establishment of the optimization model of affine registration of 3D point cloud based on pseudo-Huber loss function,the independent component analysis(ICA)algorithm is adopted to solve the initial value of the affine transformation.Then,the affine transformation is optimized by the framework of affine iterative closest point algorithm,and the optimal affine matrix A and translation vector t can be achieved.The experimental results show that,the proposed algorithm can yield high accuracy of affine registration of 3D point cloud,and the influence of noise and missing data on the accuracy of affine registration of 3D point cloud can be effectively suppressed.