基于机载LiDAR点云的道路提取算法研究

A study of methods for road extraction from airborne LiDAR data

提出一种对点云特征信息进行聚类的方法,以提取机载LiDAR数据中的道路。通过采用软件ENVI 5.3反复建立三角网实现点云滤波获取地面点云,且采用零—均值标准化对地面点云进行标准化,以消除其量纲。然后进一步利用K-means++方法对点云三维坐标聚类实现点云分割,以获取包含道路点云的类别,且对该类别中点云的高度信息进行聚类以提取道路点云。以荒漠植被区机载LiDAR为研究区,对比直接对点云高度信息聚类的结果表明:在设置相同聚类参数的基础上,直接进行高度聚类的SSE总和为2550.714,所提出的先分割后聚类方法获取的SSE总和为73.696,比直接进行高度聚类的SSE总和低2477.018,说明本方法使K-means++性能更好。对比运算速度发现,虽然采用该方法聚类消耗时间比直接聚类消耗时间多16 s,但提取结果更好,可去除非道路点云3673个。

This paper uses unmanned aerial vehicle(UAV)to load light detection and ranging(LiDAR)VUX-1 and obtained original point cloud data.After getting data,it uses Riegl LMS series software to process point clouds.On the basis of processing,it intercepts test area point clouds as study data in order to verify feasibility of this paper’s methods.The total number of study original point clouds is 272,493.And then,a method for clustering using point cloud feature information is proposed to realize the extraction of airborne LiDAR data roads.Firstly,it uses software ENVI 5.3 to achieve the point cloud by repeatedly establishing a triangulation network.Then,it uses software Matlab extracted ground point cloud three-dimensional coordinates and echo intensity of point cloud information.Of course,it also uses the Python language to program for achieving methods,which are zero-mean standardization and K-means++clustering.Secondly,it uses the K-means++method to achieve point cloud segmentation by three-dimensional point cloud clustering.After that,it obtains the category containing the road point cloud and clustered point cloud height information of this category to extract road point clouds.This paper compares direct clustering of point cloud height information.Results show that after setting the same clustering parameters,the sum of the SSEs that are directly clustered is 2557.714,and the sum of the SSEs obtained by the segmentation and clustering method proposed in this paper is 73.696,which is lower than the total SSE sum of the directly height clustered 2477.018.It means that paper method can make the performance of K-means++better.Comparing the speed of operation,it is found that,although the clustering time consumed by paper method is 16s longer than that of direct height clustering,the extraction result is better and the non-road point cloud 3637 can be removed.