Research on Point Cloud Hidden Danger Identification Method for Railway Surrounding Environment Inspection

With the rapid development of high-speed railroad, the safety inspection along the railroad also has higher requirements. In view of this, this paper focuses on the application of three-dimensional point cloud technology with LiDAR as the core in the identification of hidden dangers in railroad environment.The article investigates the application of deep learning in point cloud target detection, focusing on comparing and analyzing three mainstream technical routes: the point-based method represented by PointRCNN, the voxel-based method represented by VoxelNet, and the PV-RCNN model that integrates the advantages of both. In order to verify the model performance, this study uses the internationally used KITTI benchmark dataset of automated driving for experiments, which has a high degree of commonality between the vehicles, pedestrians, and other targets in the dataset and the typical hazards in the railroad surroundings. The experimental results show that the average detection accuracies of the PV-RCNN model in the three scenarios are 90%, 87%, and 86%, respectively, which are better than PointRCNN and VoxelNet, while its detection speed of 18.6 FPS meets the needs of most of the railroad active safety warning tasks, demonstrating the highest comprehensive application value. As a combination of high accuracy and PV-RCNN, as a detection framework that combines high accuracy and high efficiency, provides an effective technological path for solving the problem of real-time perception of safety hazards in the railroad surrounding environment.