基于激光点云数据的树木枝叶分割和三维重建

针对目前树木点云三维重建方法仅仅实现树木三维形态结构的重建，而忽视了树木表面纹理和细节的问题，本研究提出了一种超体素分割和基于kd-tree纹理映射的贪婪投影三角化算法相结合的树木点云重建方法。首先根据点云的颜色、法线和空间距离等特征，利用体素云连通性分割算法生成超体素块；然后基于不同超体素块之间的凹凸关系，利用局部凸包连接算法对点云进行聚类分割，从而实现枝干和树叶点云分割；最后利用贪婪投影三角化算法分别对枝干和树叶点云进行三维重建，并基于点云RGB信息，利用kd-tree算法对枝干和树叶模型添加真实的纹理色彩。结果表明，树木点云重建方法能实现枝干表面和树叶等细节模型的构建，还原真实的树木形状结构和纹理色彩；通过对不同抽稀层次下的点云进行测试，发现核桃楸树木模型主要枝干结构得到了很好的保留，这说明建模算法对点密度的变化具有鲁棒性。

Segmentation and 3D Reconstruction of Tree Branches and Leaves Based on Laser Point Cloud Data

Currently,three-dimensional (3-D) reconstruction of tree models based on laser point cloud data was only realized from the aspect of tree morphological structure,tree surface texture and details were neglected.This study proposed a super-voxel segmentation and greedy projection triangulation method based on kd-tree texture mapping algorithm for combining tree point cloud reconstruction.Firstly,based on the color,normal and spatial distance of the point cloud,the super-body block was generated by the voxel cloud connectivity segmentation algorithm.Then,based on the concave-convex relationship between different super-body blocks,the local convex hull connection algorithm was used to solve the clustering and segmentation of point clouds and the point cloud segmentation of branches and leaves were realized.Finally,the greedy projection triangulation algorithm was used to reconstruct the branch and leaf point clouds,respectively.Based on the point cloud RGB information,the kd-tree algorithm was used to add true texture color to the branch and leaf models.The experimental results showed that the tree point cloud reconstruction method proposed in this study could realize the construction of the details of the branch surface and leaf models,and restore the true tree shape structure and texture color.By testing the point clouds under different thinning levels,it was found that the main branch structure of the Juglans mandshurica model was well preserved,which indicated that the proposed modeling algorithm was robust to the change of point density.