Spatial‐Relation‐Guided Intelligent Understanding of Complex Railway Tunnel Structures Using Mobile Laser Scanning Data

ABSTRACT Intelligent understanding of the complex structures of railway tunnels is a crucial prerequisite for 3D urban underground space modeling and spatiotemporal pattern mining. However, most existing digitization methods still suffer from low accuracy and incompleteness, particularly in tunnel environments characterized by large volumes, complex structures, and numerous facilities. In this study, we propose a spatial‐relation‐guided intelligent method for understanding complex tunnels using mobile laser scanning (MLS) data. First, we developed a set of rules that describe the spatial relationships and geometric features of tunnel objects, forming a semantic knowledge base for classifying tunnel facilities. Next, we utilized the spatial and attribute information of points to establish constraint methods and statistical rules, including efficient model fitting and high‐dimensional spatial clustering. Finally, the 3D tunnel scene was hierarchically divided into nine categories: ground, pavement, tracks, linings, refuges, power lines, pipelines, power line support facilities, and other auxiliary facilities. The proposed method was validated on four distinct tunnel datasets. The experimental results are remarkable, with a global point‐wise accuracy exceeding 93.3%. These results demonstrate that the proposed method can effectively and accurately understand raw tunnel point cloud data, facilitating efficient management and operation of underground spaces at both macro and micro levels.