Research On Tree Point Cloud Registration Based On Fast Pseudo Feature Point

The construction of digital plants by 3D reconstruction technology is a research hotspot in the fields of graphics and agriculture.Due to the limited view range of the scanning device and the self-occlusion of the plant,3D point cloud registration has become an urgent problem in reconstructing digital plants.Tree is an important component of plants and plays an important role in China's economic and ecological development.However,there are many branches and leaves,which block each other,and therefore the point clouds captured from different angles are incomplete with low overlapping rate and certain amount of noise.The existing tree point cloud registration algorithm are less robust and efficient.At the same time,there are relatively few studies on tree point cloud registration.Based on those,a new tree point cloud registration method based on fast pseudo feature points is proposed.This method plays a key role in optimizing tree reconstruction and visualization to a certain extent.It includes initial registration and fine registration.The main contents of this study are as follows:(1)Tree point cloud acquisition and preprocessing scheme was completed.The point clouds of apple tree,magnolia tree and red maple were captured through Kinect 2.0 from multiple angles.Then background information,noise and outliers existing in point cloud data were removed through pass through filter and statistical filter,and the average removal rate is about 37.39%.Preprocess procedure provides suitable data for research on tree point cloud registration.(2)Aiming at improving extraction efficiency of tree point cloud pseudo feature point,a fast extracting algorithm of pseudo feature points was proposed.According to the spatial mor-phological structure,one tree model was divided into cluster by steps of horizontal clustering,DBSCAN clustering.Then pseudo-feature points were computed.It is fast and genetates less points with key point cloud accurately extracted.Experimental results show that the method of extracting pseudo-feature points is fast which increases by 91.46%with 79.36%less points compared with the existing method.(3)A tree point cloud registration method based on fast pseudo feature points was pro-posed.On the problem of low registration efficiency,the initial registration is based on the sparse pseudo-feature point set.Then three pairs of matching points were used to calculate the transformation matrix.KD-tree was used to search for the corresponding point pairs to further improve searching efficiency.On the problem of low registration robustness,the clustering constraint and the neighborhood angle constraint were used in matching pairs to expand the angle range of the registration.In order to improve the robustness of the registration,DB-SCAN was used for secondary clustering for the internal noise existing in the point cloud.And Tukey function was used to optimize the objective function to further improve the registration robustness.(4)Registration performance test was carried out by using the actual acquired tree point cloud.The experimental results show that the method has good registration effect under tree point cloud with leaves,no leaves,outliers and multi-angle difference.The efficiency of proposed method increased by 37.16%,99.29%respectively compared with improved SICP and CPD algorithm,and the accuracy increased by 31.07%compared with IRLS-ICP.At the same time,the method extends the registration angle range to 180~?,and the robustness is improved.