Multi View Registration Of 3D Color Point Cloud Data

With the continuous development and progress of computer technology,it is very important to obtain the complete three-dimensional color scene in the actual scene and carry on the virtual reproduction.The appearance of 3D laser acquisition technology has played a very important starting role in the study of this field.Three-dimensional laser acquisition technology in target recognition,terrain survey,industrial measurement and other fields plays a very important role.However,due to the limited view of the actual scene scanning and the occlusion of objects in the use of 3D laser acquisition system,the whole object can not be scanned in the actual scene at the same angle in the same view.Three-dimensional color point cloud data,so it is necessary to scan the object many times from different perspectives,and then transform the 3D color point cloud data from multiple angles to the same coordinate system to realize the registration of the point cloud,so as to realize the recurrence of the large scene.However,because the registration technology of three dimensional color point cloud data is not mature enough,many registration algorithms have the disadvantages of slow registration speed and low registration accuracy.In view of these problems,the following work has been done in the process of completing the thesis:The three dimensional color point cloud data acquisition system is studied deeply,and the use method of the 3D color point cloud data acquisition system is mastered.The threedimensional color point cloud data are obtained by scanning the actual scene in multiple perspectives.Then,in this paper,several methods of multi view registration for threedimensional color point cloud data are studied.Two registration methods are proposed,which are 3D color point cloud registration based on point cloud information and registration of threedimensional color point cloud based on image information.Based on VS platform,the effect of three dimensional color point cloud data registration before and after registration is designed.Software platform for monitoring registration speed and registration accuracy.Three dimensional color point cloud registration based on point cloud information,in order to solve the rotation translation matrix of two 3D point cloud scenes to be registered,first of all,we treat the two point cloud sets of the registration to calculate the geometric features and extract the point and line special collection respectively.It is necessary to select the appropriate feature point set according to the structure of the scanned scene,and then construct the Kd-tree In order to find the nearest neighbor point more efficiently and find the corresponding point after the two feature point set,the SVD(singular value decomposition)method is used to calculate the rotation translation matrix which satisfies the minimum mean distance of the corresponding point.At last,two three-dimensional color point cloud data are registered by the solution of the rotational translation matrix to the same one.In a coordinate system.On the basis of the geometric information of 3D point cloud data,it is proved that the method of extracting feature points and using feature point set is used in this paper.The time efficiency and registration accuracy are better than the ICP and 3D-NDT methods in the PCl library.The three-dimensional color point cloud registration based on image information is used to generate two-dimensional color images based on the collected three-dimensional color point cloud data.The feature calculation and feature extraction of two color images are carried out by using the SIFT(scale invariant feature transform)descriptor.Then,in the process of feature matching,RANSAC(random sampling agreement)is combined.The algorithm is used to match and correct the mismatch points in the matching process,and get the feature points corresponding to one one in the two images,and solve the spatial transformation in the image to complete the image registration.At the same time,after the matching of the feature points of the image,the corresponding feature points of the image are mapped to the corresponding points in the three-dimensional space according to the mapping relationship between the pixels of the two-dimensional image and the three-dimensional space point.The rotation transformation matrix between the two points is calculated by using the geometric information of the corresponding points in the three-dimensional space.Finally,the solved rotating flat is used.The moving matrix is used to register two 3D color point clouds into the same coordinate system to realize multi view registration of 3D color point cloud data.Experiments show that this method achieves high accuracy of multi view registration for 3D color point cloud data.