Abstract: Planar objects detection is the key approach for LIDAR data and 3D reconstruction now. Basic building types are analysed. k-d tree is used to create adjacency information of points; randomized Hough Transform is used to detect and fit the planar objects; region growing approach based on geometrical similarity and colorimetrical similarity is presented. The segmentation algorithm requires a small number of manually set parameters which are used to keep balance between under- and over-segmentation. The experiments of the presented algorithm on a point cloud of architecture show its effectiveness.

Key words: point cloud segmentation, k-d tree, Randomized Hough Transform, region growing

摘要： 从激光点云中提取建筑物平面区域是当前建筑物三维建模的关键技术。通过分析激光点云数据中建筑物的特征，引入k-d tree数据结构及随机霍夫变换（RHT）算法，提出了融合几何与颜色相似信息的区域生长分割算法。为了避免分割过程中的过度分割和分割不足，算法需要人工设置几个参数。通过在一组建筑物中提取实验数据，验证了该算法的有效性。

关键词: 点云分割, k-d 树, 随机霍夫变换（RHT）, 区域生长