An improved 3D point cloud instance segmentation method for overhead catenary height detection

The function of a subway-tunnel catenary is to ensure the safe operation of a subway under highspeed conditions; hence, it plays an irreplaceable role in the subway system. However, owing to bad tunnel environments and repeated vibrations, catenaries can easily become deformed. To solve the above problems, this study proposes a new detection method, which applies the 3D-BoNet instance-segmentation model with a multi-scale grouping (MSG) structure and transfer learning to a 3D point-cloud tunnel dataset. Experiments show that the improved model can effectively segment the left and right tracks and the catenary of a tunnel. A comparative experiment shows that the proposed method improves the average accuracy (mPrec) by 3.8% under the classical index with an intersection-over-union (IoU) threshold of 0.5. Moreover, the computational efficiency is improved by 33.01%. This method has broad application prospects in the research field of 3D point-cloud instance segmentation.

Automated summary

This study proposes a new detection method using the 3D-BoNet instance-segmentation model with a multi-scale grouping structure and transfer learning. It effectively segments the left and right tracks and the catenary of a tunnel in 3D point-cloud data. Experimental results show improved accuracy and computational efficiency.