Three-dimensional Model Of Individual Tree Based On Terrestrial LiDAR Scanner Point Cloud Data

Vegetation is the most complex ecosystem on earth and the accurate reconstruction of its three-dimensional model is of great significance to the study of carbon and nitrogen cycle,leaf area index and volume estimation of vegetation.As a new surveying and mapping technology,the terrestrial lidar has the characteristics of small spot,high efficiency,strong penetrability,high density,etc.,which not only provides a new means for the accurate acquisition of the tree point cloud data,but also provides important data support for the sense of reality and high-precision modeling of trees.How to use the terrestrial lidar data to construct the tree three-dimensional model quickly and accurately has become a research hotspot.In this paper,a Magnolia grandiflora and a Cinnamomum camphor were taken as the research objects,and the branch point cloud extraction method and a high-precision three-dimensional tree reconstruction technique were proposed based on the point cloud data obtained by terrestrial lidar.The main research works and conclusions are as follows:?1?Branch point cloud extraction algorithmAccording to the distribution of tree point cloud,the distribution of point cloud at branches is roughly cylindrical,and the point cloud at leaves is approximately planar.A new algorithm for the branch point cloud extraction was proposed by combining classification and segmentation.First,the tree point clouds data were sliced horizontally along tree height direction using the axial bounding box algorithm.Then,the K-means clustering method and random sampling consistency?RANSAC?algorithm were used to divide the canopy point cloud into three categories:the wood point cloud,the leaf point cloud and wood and leaf mixed point cloud.Finally,the cylindrical segmentation method based on the RANSAC algorithm was used for the precise extraction of wood component in wood and leaf mixed point clouds to obtain the accurate extraction of branches point clouds,so as to realize the separation of branches point clouds from the leaf point clouds.The results were compared with those obtained using the cylindrical segmentation method based on the RANSAC algorithm alone.The performance of the algorithm was tested using recall?Re?and precision?Pr?.The average recall and precision in extracting canopy wood point clouds of the Magnolia grandiflora and Cinnamomum camphor achieved 94.60%,92.02%and 93.62%,91.46%,respectively.The proposed method had higher accuracy in extracting wood components compared to the cylindrical segmentation method alone.?2?Extraction of the parameters of tree growth and skeleton points of branchesThe tree height,DBH and crown width were extracted by the fitting circle algorithm based on least square method and convex hull algorithm.Since the skeleton point of the tree is located at the center of the tree branch,which can effectively represent the geometric or topological features of the branch,the L₁-median algorithm was used to extract the skeleton point of the tree.The density weight method was used to optimize the skeleton point so as to obtain the skeleton point of branches.?3?Tree three-dimensional model constructionCombining the skeleton points which were extracted by L₁-median algorithm and the radius information of the corresponding skeleton points which was obtained by using the least square fitting circle algorithm,the generalized cylinder was drawn by OpenGL technology to build a three-dimensional model of branches.Since the leaf has a different geometry from the branch and the amount of single leaf data is small,Firstly,the leaf point cloud data was manually segmented into independent single leaves.Then the least square method was used to fit the surface of the leaf point cloud data,so as to accurately extract the complex contour information of leaves.Finally,the three-dimensional model of the leaf was constructed by using the Delaunay triangulation algorithm.