A 3D LiDAR-Based Approach for Dynamic Human Detection and Intrusion Determination in Coal Conveyor Galleries*

Abstract

To meet personnel safety monitoring demands in coal conveyor galleries and overcome traditional 3D monitoring limitations, a 3D LiDAR-based dynamic personnel detection and boundary-crossing judgment method is proposed. Utilizing 3D Li-DAR as the sensing terminal, this method constructs a dual-layer temporal sequence detection architecture as the first innovation, comprising point-wise microsecond-level initial detection and frame-level occlusion-accumulation refined detection to precisely extract complete dynamic personnel point clouds. For region calibration, an interactive spatial mapping mechanism based on raycasting and hardware-accelerated rendering is proposed, achieving high-fidelity and efficient calibration of 3D bounding boxes for unstructured hazardous areas. Regarding intrusion behavior judgment, a multi-variable adaptive threshold function coupling detection distance, target relative pitch angle, and horizontal azimuth angle is established as the second innovation for accurate recognition. Cross-scene testing on public benchmark datasets (KITTI, Apollo, and CARLA) demonstrates a maximum frame rate of 24.2 FPS and a 41.3 ms average latency. Furthermore, on-site industrial testing confirms that under dust and complex lighting interferences, the system achieves a 90.02% average detection accuracy for various boundary-crossing behaviors with a 33.6 ms per-frame processing time, fully satisfying real-time, high-precision industrial monitoring demands.