Registration Algorithm Of Plant Point Cloud Based On Pseudo Feature Point

In order to get the complete three-dimensional(3D)point cloud model from tree point cloud quickly,point cloud registration is essential.At present,experts and scholars put forward many point cloud registration algorithms,but the surface of tree is rough,branches are slender and have serious self-shielding,3D scanner only get incomplete point clouds and theses point clouds contain noise,existing algorithms cannot fully adapt to the unique characteristics of tree.Based on this,this paper puts forward a registration algorithm based on pseudo feature points;this algorithm is divided into the initial registration and fine registration.The main research contents and innovations of this paper are as follows:(1)Propose the extraction algorithm of pseudo feature points.Aiming at this problem that the structure of tree was complex and it was difficult to extract feature points.This paper used the extraction algorithm of pseudo feature points to extract pseudo feature points for tree point cloud,this method finished the extration of pesudo feature points by clustering,computing representive points,and got a better set of pseudo feature point.Compared with the feature points extraction algorithm based on geometric features,the extration of pseudo feature points is more suitable for the feature of tree point cloud.(2)Propose a new registration based on pseudo feature point.For this problem that point cloud was dense,registration must consume lots of time,this paper used feature points roughly to adjust the location of two pieces point cloud in initial registration,reducing the number of iterations in fine registration and improving the efficiency of registration.In order to solve this problem that the method of fining corresponding point pairs is sensitive to noise.This paper used the similarity of the neighborhood information distribution to delete error corresponding point pairs to improve the correct rate of corresponding points.In addition,According to different characteristics of point cloud data used in initial registration and fine registration,this paper used the angle and distance to measure the similarity of the neighborhood distribution to improve the accuracy of corresponding points respectively.(3)In order to verify the registration algorithm based on pseudo feature points,this paper used non-leaf and leafy tree point clouds to verify the validity of proposed algorithm;and used non-plant of tree point cloud to verify the scalability of this algorithm;At the same time,under the same of experiment environment and experimental data,compared with other registration algorithms,and verified the superiority of this algorithm.Experimental results show that underthe same number of iterations,compared with ICP(Iterative Closed Point)algorithm,registration error of proposed algorithm is reduced by 41.1%,and compared with the sparse ICP algorithm,the registration error is reduced by 16.8%.In addition,this paper also used potted model,bunny to verify the versatility of algorithm.Experimental results show that this algorithm can register non-plant point cloud,having a strong commonality.