Automatic Reconstruction Of 3D Geometrical Tree Model Based On Point Cloud Data

Virtual plant has been considered as a potentially effective tool,which is beneficial to the understanding,expression and simulation of plant structure and growth processes.The three-dimensional(3D)sampling points,or point cloud data of the scanned object surface can be acquired automatically,quickly,accurately and in high density by using the 3D laser scan technology with the premise of non-destructive power.This technology has great potential for the applications such as agriculture and forestry resources inventory.The research on 3D tree model reconstruction based on range data has become one of the focuses in the field of virtual forest environment or virtual geographic environment.On the basis of software PC2Tree,which was developed by our research group,an improved method for tree branching structure reconstruction based on point cloud was proposed based on a variety of algorithms.The main research topics and achievements of this thesis are as follows.(1)Through the analysis and comparison of the existing computer plants(or trees)modeling methods and an overview of the three-dimensional reconstruction of tree point cloud,advantages of 3D model reconstruction based on point cloud is summarized.(2)In order to deal with the complex shapes of trees,point cloud from the branches and leaves was separated by using both the similarity of principal direction between neighboring points and the distribution density of points.Firstly,normal vector of point clouds were estimated by the principal component analysis(PCA).Then the principal direction at each point was estimated with chord and normal vectors(CAN).In the part of branches,the distributions of these points are in cylinder-shaped pattern,and principal directions of local points are identical to the axis of cylinder.On the other hand,point cloud dataset in the part of leaves are volumetric distributed and the principal directions varied to a great extent.According to the distribution of the point cloud and the rule of principal direction of different parts between branches and leaves,point cloud segmentation was realized.(3)Using different methods to extract the skeleton points at different positions of the tree,and then combine them to form the three-dimensional model.Skeleton points extraction procedure for stems is as follows:firstly,main branches were divided into several level sets based on the level set method;secondly,each level set points were projected onto the relevant optimal cutting plane;then on this plane circle was fitted using least square method based on the projected points;for the fitting result,center of the circle is the skeleton point,and the radius of the circle will be the branch radius.For the leaves,the crown volume was divided into equal-sized voxels,all the points in a voxel were represented by the voxel’s centroid,and all centroid points formed feature points of the crown,namely skeletons of the canopy.Results showed that the reconstructed branches model based on extracted skeleton points were in good agreement with the original point cloud,and could be used to express the shape of the trunk.The canopy feature points could express the convex and concave characteristics of the canopy contour completely.(4)From the reconstructed three-dimensional tree model,tree height,crown width,diameter at breast height(DBH),volume and other parameters were extracted,and the accuracy of the model was analyzed by visual inspection and quantitative analysis.Visual inspection is done by superimposing the branch model on the point clouds of branch,and superimposing the final model on the original point cloud,which can visually check the degree of coincidence between the model and the point cloud.The results show that the error ranges of tree height,crown width and DBH are 3.02%~5.05%,2.57%~3.63%,2.83%~8.05%,respectively.For the tree volume we only compared our result with that from the third-party software SimpleTree,so the accuracy of which needs further study.(5)Some modifications were made for the PC2Tree software,including the separation of tree branch and leaf point cloud,and the radius calculation method.The revised PC2Tree is capable of 3D reconstruction of trees with higher accuracy and greater fidelity.