Safety optimization of grab unloaders based on machine vision and 3D coordinate system reconstruction

With the increasing complexity and automation of port logistics systems, ensuring the operational safety and real-time perception capability of ship unloaders has become a pressing issue in intelligent port equipment research. This study proposes a machine vision-based method for dynamic 3D coordinate system construction to optimize safety during unloader operations. A comprehensive system integrating RGB-D cameras, LiDAR sensors, IMU units, and SLAM algorithms is designed to achieve real-time 3D mapping and risk perception. The point cloud environment is enhanced through Voronoi-based grab bucket skeleton recognition and trajectory prediction. Experimental validation was conducted in laboratory and actual port environments. The 3D mapping module achieved a maximum root mean square error (RMSE) of 6.1 cm and a real-time frame rate of 14 FPS under night conditions. Collision prediction accuracy reached 91.2% with an average response time of 1.4 s. Over a 7-day continuous test, only 3 false alarms were recorded out of 68 warning events, confirming the system's robustness and reliability. The results demonstrate that the proposed method significantly enhances spatial awareness and operational safety of unloaders, offering potential for broader applications in large-scale port equipment automation.