Research On The Method Of Building Single Tree Branch Skeleton Based On Point Cloud Data

The topological structure and geometric shape features of the plant model can be reflected by the skeleton,which is the basic element of three-dimensional model matching,retrieval and surface reconstruction.So,it is widely used in the field of plant modeling.At present,the establishment of plant models includes image-based modeling method,process modeling method based on plant growth rules,and modeling method based on artificial sketch.Digital close-range photogrammetry is used in the field of forestry measurement and tree modeling.Plant models can be captured,but the tree growth and geometric parameters cannot.The rule-based approach requires some botany expertise to define the three-dimensional shape of a plant by defining and adjusting geometric parameters.Based on the artificial sketch method,virtual plants with vivid appearance and various shapes can be generated.With the continuous improvement of three-dimensional laser scanning technology in recent decades,the acquisition speed and accuracy of point cloud data on the object surface are rapidly increasing.The three-dimensional laser scanning technology has the characteristics of automatic and rapid measurement without direct contact with the object.Higher density and accuracy are the advantages of real-time measured data.Obtaining large-scale three-dimensional points on the tree surface with the three-dimensional laser scanning of tree plants,measuring the geometry of tree growth and extracting its physiological parameters are hot spots in the forestry field in recent years.The focus of this paper is to build a single tree skeleton based on the point cloud data of tree branches.The main work of this article is summarized as follows:(1)The three-dimensional laser scanning technology was explored from the principle of scanning and the characteristics of the scanning technology.The precautions in the process of tree data collection were summarized with laser scanner.The method of organizing and retrieving 3D point cloud data were mastered,and Kd-tree structure to store point cloud data were used.(2)The method of precise registration of point cloud data was mastered,and the initial denoising algorithm of point cloud data was studied.Laplacian denoising,mean filtering denoising,bilateral filtering denoising,and least squares denoising were experimentally compared.Based on the de-noising of the Laplace algorithm,parameter adjustments are made to achieve initial de-noising.The de-noising of the layered data is performed through the spatial point neighborhood relationship and the standard deviation of the mean value,and the de-noising is integrated twice to achieve a better de-noising effect,which can better protect the characteristics of the model and reduce the model.The resulting deformation and wear,the edges are becoming more and more clear.(3)A method for constructing the trunk skeletons of tree plants was designed and implemented.The tree branch point cloud was segmented and de-noised twice.The K-means++ clustering method was used to distinguish the different branches and calculate the skeleton points.The minimum spanning tree algorithm was used to connect the skeleton points of different layers and finally constructed.Compared with previous three-dimensional point cloud clustering,segmented point cloud clustering is able to find the skeleton skeleton points more accurately,to obtain a more accurate skeleton model.