基于FCM的机载LIDAR数据的建筑物和植被分类方法  

Classification method for construction and vegetable of airborne LIDAR data based on FCM clustering arithmetic

作　　者：刘昌军[1] 黄乾[2] 唐瑜[3] 卢雨平 

机构地区：[1]中国水利水电科学研究院,北京100038 [2]山东省水利科学研究院,山东济南250013 [3]中水顾问集团成都勘测设计研究院,四川成都610072

出　　处：《中国水利水电科学研究院学报》2013年第3期189-194,200,共7页Journal of China Institute of Water Resources and Hydropower Research

基　　金：国家国际科技合作计划资助(2010DFA74520);"十一五"国家科技支撑计划项目(2008bab42b05;2008BAB42B06)

摘　　要：针对LIDAR点云数据中建筑物和植被难以快速分类的问题,提出了应用FCM(Fuzzy C-Mean)模糊聚类的方法对离散机载激光点云数据进行建筑物和植被分类的方法。首先针对机载点云数据的特点采用了平面投影的De launay构网方法进行点云的三角网重构,然后根据三角网的法向矢量信息的属性不同,利用FCM方法和改进的方位矩阵方法对其进行模糊聚类,进而实现建筑物和植被等不同属性的点云分类。该方法可快速将点云进行分类,且分类结果可用不同颜色进行空间显示。在此基础上,采用IDL(Interface description language)语言编制了三维激光点云可视化分类软件LIDARVIEW。并应用该软件对某区域的机载点云数据进行了分类实验。实验结果表明:(1)基于平面投影的Delaunay构网方法特别适合机载LIDAR点云数据的快速构网,且该方法构网速度快、效率高;(2)应用FCM模糊群聚的方法和改进的方位矩阵方法适用于机载LIDAR数据的植被和建筑物分类,分类速度快且效果好;(3)FCM模糊群聚方法对机载LIDAR数据的群聚分类结果可靠、合理,具有较强的通用性和推广性。Aiming at the problem that quick classification of buildings and vegetation is difficult to achieve for Lidar point clouds, the application of fuzzy clustering method FCM （fuzzy c-means） to the classification of buildings and vegetation for the discrete airborne laser point clouds is proposed.First of all, triangulation reconstruction by Delaunay triangulation based on planar projection is carried out according to the character- istics of the point clouds. Then, according to the different properties of the normal vector, fussy clustering is conducted by FCM and the improved orientation matrix method. Furthermore, the point clouds classifica- tion for different properties such as buildings and vegetation is achieved. This method can quickly classify point clouds and the classification results can be visualized in different colors in space. On this basis, soft- ware for three-dimensional visualization of the laser point clouds classification is developed with IDL lan- guage and named LIDARVIEW. With this software, airborne point clouds in one region were selected for the data classification experiments. The experimental results show that： （1） Delaunay Triangulation based on planar projection is particularly suitable for the rapid TIN （Triangulated irregular network） construction of airborne LIDAR point clouds, having the advantage of faster speed and high efficiency; （2） Applica- tion of the fuzzy clustering method （FCM） and the improved orientation matrix method is suitable for vegeta- tion and buildings classification for airborne Lidar point clouds, being fast and effective; （3） Results for airborne Lidar point clouds by FCM are reliable and reasonable, with a strong versatility and generalization.

关 键 词：机载LIDAR数据 点云分类 FCM 模糊群聚 改进的方位矩阵 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]