Abstract Truck cranes overturning accidents pose significant safety risks of railway infrastructures and may result in severe structural damage, highlighting the need for overturning supervision systems. To address the challenges of online overturning supervision for truck cranes, the paper proposes a key parameter identification framework of truck crane using LiDAR for real‐time remote monitoring. First, it proposes a truck crane segmentation network (TCSegNet), a novel semantic segmentation model integrating graph convolution and shape‐aware attention mechanisms, outperforming benchmark methods in point cloud semantic segmentation of critical crane components (e.g., chassis, boom, and slewing platform) in complex scenarios. Building upon precise segmentation, it develops a geometrically rigorous parameter identification pipeline based on multi‐state plane fitting and point cloud registration of truck crane under different operation states, enabling accurate estimation of stability‐critical parameters including boom yaw angle (error < 1.3°) and pitch angle (error < 2.6°). To train and evaluate the proposed approach, a truck crane point cloud dataset including both simulated and real‐world data is created. Experimental comparison with the state‐of‐the‐art methods on the truck crane dataset shows the outperformance of the TCSegNet model. The proposed approach provides high‐precision 3D monitoring indicators for overturning risk assessment, significantly enhancing safety supervision capabilities of truck crane.
Sun et al. (Wed,) studied this question.