A New Registration Method Of Point Cloud Data

Three-dimensional reconstruction technology is a significant part in computervision. The3D reconstruction of real object has a very broad prospect of application inevery field of life, including industrial automation, archaeology, cultural heritageprotection and computer aided medical. With the increasing maturity of the3D scanningterminal, there exists no troublesome problem in getting accurate partial3D modelcoordinates and a good registration algorithm for3D point cloud data is graduallybecoming the key to the3D reconstruction technology. Therefore the research onregistration algorithm has a very important significance to three-dimensional digitalprocessing technology.This paper focuses on3D point clouds registration algorithms in the pipeline of3D reconstruction and presents a novel and high-accuracy registration algorithmwithout ICP method. The proposed registration method can align pairs of point cloudsin arbitrary initial position robustly from rough to fine using congruent-spheres-basedpropagating algorithm.The proposed congruent-spheres-based propagating algorithm is a novel methodfor registration. The algorithm combines the sphere rotation invariance with theorthogonal force consistency algorithm creatively to overcome the limitation of usingone of each when it comes to the search for corresponding points. The correspondingmethod can obtain the high accuracy of corresponding points and lay a solid foundationfor registration work. As to the fact that a good registration algorithm must have aglobal convergence, a propagating framework based on the congruent spheres isproposed to get corresponding points spreading all over the point cloud data. In theend, the RANSAC algorithm is employed to exclude some false correspondences inthem, and the final rigid motion between the pair wise point clouds is calculated in aleast-squares sense by the leaving reliable correspondences. Experimental resultsdemonstrate our method is not only robust, but finer than the traditional ICP algorithm.Further, this scheme can be utilized to examine the symmetry of the point clouds.