STSD:A large-scale benchmark for semantic segmentation of subway tunnel point cloud

Deep learning (DL) semantic segmentation of tunnel point cloud shows an efficient path for applications related to subway tunnel scenes, such as health inspection and building information modelling (BIM). Current methods for tunnel point cloud segmentation often suffer from a shortage of benchmarks. This paper proposed a large-scale, multi-modal dataset for semantic segmentation of subway tunnel point cloud called subway tunnel segmentation dataset (STSD). The STSD comprises point clouds and projected images annotated into 12 categories, encompassing three types of subway tunnels with a combined length exceeding 2700 m, totaling over 2.26 billion points. A novel approach for DL semantic segmentation of subway tunnel point clouds is proposed herein. This approach enables the direct utilization of image-based DL segmentation networks on subway tunnel point clouds. Furthermore, it incorporates a lossless coordinate transformation method capable of converting tunnel point clouds of any cross-section shape into images with minimal information loss. Further evaluation of several classic or state-of-the-art 2D and 3D DL semantic segmentation models shows the feasibility of the approach and dataset. The best 2D model achieves a mIoU of 86.26% and outperforms the best 3D model by almost 10%. This research provides a novel approach for DL semantic segmentation in subway tunnel point clouds, contributes a large-scale, multi-modal dataset for the tunnel semantic segmentation, and creates a benchmark for further evaluation of the corresponding algorithms.