顾及上下文信息的城区机载LiDAR建筑物自动化提取方法研究  

Contextual Extraction of Urban Buildings from Airborne LiDAR Point Cloud

作　　者：陈宇琪 罗一芬 龙竞帅 郭雪莹 CHEN Yuqi;LUO Yifen;LONG Jingshuai;GUO Xueying(Haikou Land Surveying and Mapping Institute,Haikou 570100,China;Hainan Hydrogeological Engineering Geological Survey Institute,Haikou 570102,China;CNBM Geological Engineering Exploration Academy Co.,Ltd.,Haikou 571127,China)

机构地区：[1]海口市土地测绘院,海南海口570100 [2]海南水文地质工程地质勘察院,海南海口570102 [3]中材地质工程勘察研究院有限公司,海南海口571127

出　　处：《地理信息世界》2019年第5期58-63,共6页Geomatics World

摘　　要：针对从LiDAR点云数据中提取建筑物难的问题,提出一种顾及上下文信息的机载点云建筑物自动化提取方法。首先,以点为分类基元提取视觉分类特征,构建描述点云场景的视觉空间,并利用随机森林分类器初步分类场景。然后基于条件随机场模型将空间上下文信息引入点云分类中,使得分类结果满足局部连续和全局最优的特点。实验结果表明,建筑物可以有效地被分离出来,分离正确率超过96%,将为后续建筑物的自动矢量化提供理论支撑。To address the classification difficulty of distinguishing buildings from others, this paper proposes an automatic building extraction method based on context information from airborne point clouds. First, the single-point-based visual features are extracted, and the visual space describing the point cloud scene is constructed. Then, the random forest classifier is used to initially classify the scene. Finally, based on the conditional random field model, the spatial context information is introduced into the point cloud classification so that the classification results satisfy both local continuity and global optimum. Experimental results show that the buildings can be effectively separated from the others with the overall accuracy of over 96%, which is the theory support for automatic building vectorization at the subsequent steps.

关 键 词：机载LIDAR 随机森林 条件随机场 点云分类 

分 类 号：P234.4[天文地球—摄影测量与遥感]