Multi-Class 3D Tunnel Point Cloud Segmentation Using a Deep Learning Method

A deep learning method is proposed to act on point clouds for segmentation, which can feed the data into a built network based on an encoder-decoder architecture coupled with an improved 3D dual attention module to extract and learn features. To verify the effectiveness and feasibility of the proposed model, a tunnel point cloud dataset collected in a metro tunnel project is used. The experimental results show that the proposed model has a plausible performance with an Intersection over Union (MIoU) of 0.8597, and it outperforms other state-of-the-art methods such as PointNet and DGCNN. Overall, the proposed model shows excellent performance and provides effective and accurate results for multi-class segmentation on 3D tunnel point clouds.