Real-time rail recognition based on 3D point clouds

Accurate rail location is a crucial part of safety monitoring in railway operational systems. Light detection and ranging can be used to obtain point clouds that contain three-dimensional (3D) information about the railway environment, especially in darkness and poor weather conditions. In this paper, a real-time rail recognition method based on 3D point clouds is proposed to solve the challenges, such as the disorder, uneven density and large volume of the point clouds. A voxel downsampling method is first presented to balance the density of the railway point clouds, and pyramid partitioning is designed to divide the 3D scanning area into voxels with different volumes. A feature encoding module is then developed to find the nearest neighbor points and to aggregate their local geometric features for the central point. Finally, a multiscale neural network is proposed to generate the prediction results for each voxel and the rail location. Experiments are conducted on nine sequences of 3D point cloud railway data. The results show that this method has good performance in detecting straight and curved rails and other complex rail topologies.