基于网格化曲率聚类的点云分割

针对三维散乱点云数据，提出一种基于网格化曲率聚类的点云分割算法．首先对点云进行三维空间动态网格划分建立散乱数据点的拓扑关系，利用坐标转换法，在局部坐标系内拟合抛物面，进而求得高斯曲率、平均曲率等微分信息，然后基于聚类分析的思想利用高斯曲率和平均曲率的相似性进行点云数据的初始分割，再利用空间网格的拓扑关系检查和纠正完成点云数据的区域分割．该方法不需计算出每一个测量点的曲率值，从而提高了计算速度．

Segmentation for Point Cloud on the Grid-Based Clustering of Curvature

For 3D scattered point cloud data, a new method of segmentation of point cloud on the grid- based clustering of curvature is proposed. The raw point data set is first sorted into dynamic 3-D grids to build the topological relations for the point data, and by coordinate transformation method, fitting with a quadratic paraboloid with the point data in each 3D-grid to get the differential properties of the center point in the grid, such as gauss curvature, mean curvature and so on. Then based on the idea of cluster analysis using the comparability of the gauss curvature and the mean curvature the first segmentation is done, and then the first segmentation result by the spatial topological relations of the 3-D grids to finish the segmentation of point cloud data is inspected and corrected. The method does not need to calculate the curvature of the each point, so it improves the calculating speed.