Research On Scattered Point Cloud Data Matching Technology

Point clouds matching technology as a key problem in point cloud processing,it has been applied in many fields,such as reverse engineering,virtual reality,heritage,robot vision and medical technology,and this technology is the most widely used in reverse engineering,it plays an important role on point cloud surface reconstruction.Traditional point clouds matching techniques often require artificial assistance,such matching technology in many aspects of robustness is not strong,the accuracy is not high and relatively time-consuming,so fast,accurate and strong robustness point clouds automatically matching technology has long been a research focus among point cloud data processing.In this thesis,on the basis of both directions from the spin images and normal distributions transform,studying corresponding improved algorithms to achieve a feasible and effective point clouds matching technology.Firstly,the thesis studies spin image algorithm for matching point clouds,the method through by two distance relationship of one point with its surrounding points to establish spin image,which transforms the 3D information into 2D information indicated that easy to find the corresponding point.But for matching point clouds with noise,which robustness is not strong,so it is improved by using an angle information replaces one of the 2D distance information,and the improved algorithm has a strong robustness while matching point clouds with noise.Secondly,this thesis studies the point clouds matching algorithm based on the normal distribution transform,it is a simple way to describe the surface,this algorithm has a key parameter that is size of voxel,voxel is a cubic lattice and used to segmentation the point cloud,its size determines the result of point cloud matching is good or bad,and usually using a fixed size of voxel,points are uneven distribution for point cloud and this division will be unbalanced,so causing undesirable result of matching.Thus,this algorithm has been improved,and using variable size of voxel to segmentation point cloud,so the accuracy of result of point clouds matching has been improved.Finally,a lot of experiments have been done to compare the two type algorithms respectively.The results verified that the improved algorithm of spin image has stronger robustness while matching point cloud with noise than its original algorithm,and the improved algorithm of normal distributions transform also has higher precision than itsoriginal algorithm.