基于TLS数据的站场线路点云提取算法  

Point cloud extraction algorithm based on TLS data in railway stations

作　　者：方一鹏 宋占峰[1] 李军[1] FANG Yipeng;SONG Zhanfeng;LI Jun(School of Civil Engineering,Central South University,Changsha 410075,China)

机构地区：[1]中南大学土木工程学院,湖南长沙410075

出　　处：《铁道科学与工程学报》2024年第2期545-554,共10页Journal of Railway Science and Engineering

基　　金：国家重点研发计划(2021YFB2600403)。

摘　　要：铁路站场线路几何信息对于铁路安全管理与维护具有重要意义。由于铁路站场内包含多条线路,且轨道错综复杂,使得从大场景点云中自动提取多股道钢轨点云成为难题。地面激光扫描TLS(Terrestrial Laser Scanning)作为非接触式测量手段,可快速获取铁路场景中的海量点云数据。针对TLS技术获取的铁路站场点云数据,提出一种基于Delaunay三角网聚类的多股道钢轨点云提取算法。基于分割-归并的思想,在获取铁路站场高精度点云后,沿站场线路方向将点云分为若干段,基于轨道平顺性特征,利用三角网聚类算法逐段提取钢轨顶面点云。在归并阶段整合站场中各股道轨面点云信息,将各段轨面点云连接起来,同时匹配左右轨面点云。将该方法在玉林站部分站场区域进行实例验证,提取到的轨道点云在对象层面上的总体精度为93.95%,完整度为90.57%,准确度为97.59%,相较于平面格网法,提取总体精度提升了5.65%,准确度提升了18.49%。在10处截面提取轨面宽度与轨距,统计结果表明轨面宽度中误差为5.2 mm,轨距中误差为5.3 mm,满足工程精度需要。实例结果表明,算法可准确有效提取站场多股道钢轨顶面点云,为铁路场景中其他结构物的TLS数据提取工作提供借鉴思路。The geometry information pertaining to railroads within stations holds significant importance for railroad safety management and maintenance.Since there are multiple rail tracks in a station,it is difficult to automatically extract point cloud of tracks from large field point cloud.Terrestrial Laser Scanning(TLS),as a non-contact measurement means,can quickly obtain large amounts of point cloud data in railroad scenes.Based on Delaunay triangle network clustering,a point cloud extraction algorithm was proposed to extract rail track point cloud from the point cloud data of a railroad station obtained by TLS technology.Based on the idea of segmentation-merging,the method segmented the high-precision point cloud data of railway station yards into several sections along the direction of the railway tracks,and then used the triangle mesh clustering algorithm to extract the steel rail top surface point cloud based on the track regularity feature of the railway tracks.In the merging stage,the rail surface point cloud information of each section in the station yard was integrated,and the rail surface point cloud of each section were connected while matching the left and right rail surface point cloud.The method was verified using data from Yulin station,demonstrating an overall accuracy of 93.95%at the object level,the integrity of 90.57%,and an accuracy of 97.59%.Compared with the planar grid method,the overall accuracy of extraction is improved by 5.65%and the accuracy is improved by 18.49%.Based on the extracted width of trail surfaces and gauges at 10 cross sections,statistical results indicate that the standard deviations of the rail surface width and the gauge are 5.2 mm and 5.3 mm respectively,meeting the precision requirement of engineering.The example results show that the algorithm can accurately and effectively extract the point cloud from the top surface of the tracks of multiple railroads within the station,which provides a reference idea for the TLS data extraction work in other structures within railroad s

关 键 词：地面激光扫描 点云 主成分分析 DELAUNAY三角网 聚类算法 

分 类 号：U216.3[交通运输工程—道路与铁道工程]