随着激光雷达三维点云的发展，从自然环境中获得单木点云分类结果能够更好地在树木建模、植物生长模拟和遥感等领域应用。单木点云扫描结果往往只有面向扫描仪的部分，枝干结构不完整，且存在遮挡噪声等问题。因此，本文提出了一种基于骨架检测的叶干分离方法。首先对点云进行降噪处理，对较为分散且没有明显几何特征的离散点进行去除。对余下点云进行基于拉普拉斯算子收缩的骨架计算方法，获得点云骨架。主干部分使用圆柱检测的方法进行提取和补充；通过对法向量、曲率和邻近点数目的计算，对枝干细枝进行补充。得到最终的枝干及其互补叶片部分。此方法输入参数可控，同时能够较好的提取枝干部分。

With the development of Laser scanner three-dimensional(3D) point cloud, it is better for tree reconstructing, plant growth simulation and application of remote sensing when we use the segmentation results of single tree scan data from natural environment. The raw scan data usually is the part that are faced with laser, which makes the structure of branches incomplete and including noises. Therefore, we present a method for segmentation of leaves and branches based on extracting skeletal curves. First, remove the useless outliers. Then，use the remaining points to extract the curve skeleton by Laplacian. The trunk is gotten by cylinder detection, while stems are taken by normal, curvature and number of the nearest points. Finally, combine the trunk, stems and the points near the skeleton to access to branches. The complementary part of braches is leaves. Our approach’s parameters are easy-to-manipulate and extract branches effectively.