Objective In view of the low efficiency of traditional metro tunnel deformation monitoring, and the issues of existing tunnel cross-section extraction methods being susceptible to interference from auxiliary facilities and having low extraction accuracy in complex scenarios such as curved sections, a cross-section extraction method for tunnel point clouds integrating coupled segmentation of the RANSAC （random sample consensus） algorithm and connected domain with a bidirectional projection central axis constraint is proposed.

Method First, a point cloud segmentation model integrating RANSAC algorithm shape detection and connected domain analysis is constructed to achieve the refined classification and segmentation of 11 types of typical facilities, such as tunnel walls, ventilation ducts, and tracks. This effectively eliminates point cloud interference caused by various auxiliary facilities, thereby obtaining a complete and pure point cloud of the main tunnel structure. Then, a bidirectional projection algorithm is adopted to perform spatial mapping on the main tunnel point clouds to accurately extract the central axis information that can characterize the overall alignment and morphological features of the tunnel. Finally, using the extracted central axis as a reference, continuous and complete tunnel cross-section features are extracted segment by segment along the tunnel alignment.

Result & Conclusion  The test results indicate that the adopted tunnel point cloud segmentation algorithm achieves overall accuracies of 92.74% and 97.51% in straight and curved sections, respectively, and the Intersection over Union （IoU） of core facilities such as tunnel walls exceeds 0.85. The extracted central axis information can accurately reflect the alignment characteristics of the tunnel, providing a reliable reference for tunnel cross-section extraction, as well as offering high-quality basic data support for the safety monitoring of metro tunnels.