Studies On The Reduction And Registration Of 3D Point Cloud Data

With the continuous development of 3D scanning equipment,point cloud data models have been widely applied in reverse engineering,solid modeling,geographic information systems,and medical graphics and other fields.But the huge amount of three-dimensional point cloud data result in much higher requirements in the computer storage and processing.In order to alleviate the contradiction between the computer performance and the accuracy advancement of point cloud equipment,the point cloud data processing techniques have attracted becomes more and more attention.How to process three-dimensional point cloud data more accurately and more rapidly has become a hot research area for many researchers.This thesis devotes itself in the reduction and registration of point cloud data.The main work and contributions of this thesis are as follows:First,a point cloud data reduction method was proposed based on the features and the bounding box random sampling method.This method sequentially reduce the point cloud data upon randomly sampling points in the bounding boxes and implementing a reduction algorithm based on the features of points.This method can not only ensure a proper amount of data points but also ensure the homogeneity of the point distribution.Furthermore,this method can be implemented very efficiently.Second,a registration algorithm was proposed for the registration of scattered data.Different from most existing curvature based registration algorithms,this method consists of one sequential filtering process for erroneously matched point pairs and one parameter estimation process based on the Hough transform.In the filtering process,the candidate matched point pairs from different clouds are first extracted based on the curvature which are then filtered sequentially based on two similarity measures for the invariant signature and the persistence feature histogram(PFH).In the parameter estimation stage,upon parameterizing both the rotation matrices and the translation vectors of these point pairs,the Hough transform was used to remove the contributions of the erroneously matched point pairs and determine the final transform for registration.Experimental results show that our proposed algorithm can be used for the registration of partially-overlapped clouds with any relative deviation and achieve higher accuracies and robustness to noise.Finally,a method of filling holes is proposed.First,the overlapping point cloud data are removed and the feature points of the cloud boundary are extracted,after which feature points of the hole boundary are homogenized.Second,holes are filled with the partial concentric circle expanding algorithm.Finally,sparse processing is implemented with regard to the new filling points in each concentric circle.The cloud data after hole filling procedure are then used to reconstruct the surface with a method based on the Delaunay triangulation.Experimental results show that the method can repair the holes and obtain the better surface.