A Method for Extracting Main-Structure Point Clouds of Railway Overhead Contact Systems Based on Regional Geometric Representation and Cross-Dimensional Progressive Constraints
编号:162
访问权限:仅限参会人
更新:2026-10-19 18:03:45 浏览:20次
张贴报告
摘要
Abstract—An unsupervised method is proposed to extract the main-structure point cloud of a railway overhead contact system (OCS) from mobile laser scanning data in which tunnel walls, masts, and scattered background points are intermingled with slender OCS components. First, tunnels and open-air sections are partitioned using the y-direction span of longitudinal slices and the XOY convex-hull area of their lower-side points, enabling region-specific background filtering. A 10-millimeter voxel skeleton is then processed by DBSCAN, largest-XOY-convex-hull screening, and recursive three-dimensional KD-tree distance filtering to identify the main cluster, after which the labels are mapped back to the original point cloud. Tests on six files from two railway lines achieved micro-averaged precision, recall, F1 score, and intersection over union of 99.77%, 97.61%, 98.68%, and 97.40%, respectively. The retention rates of the main catenary, fine structures, and cantilevers were 99.63%, 99.81%, and 87.02%, respectively. The results show that the method suppresses complex backgrounds while retaining fine structures without training samples, thereby providing a reliable main-structure point cloud for subsequent OCS geometric-parameter computation.
关键词
Electrified railway,overhead contact system,mobile laser scanning,point cloud extraction,regional geometric representation,cross-dimensional constraints,unsupervised method
稿件作者
Yicheng Xiang
Southwest Jiaotong University
Jinfa Guan
Southwest Jiaotong University
Junqing Chen
Southwest Jiaotong University
Chenlin Xu
Southwest Jiaotong University
Jiayu Fu
Southwest Jiaotong University
发表评论