Automated UAV-based inspection of unsafe mobile elevated work platform operations using monocular 3D detection

Fall from height (FFH) remains a leading cause of fatalities in construction, while the widespread use of mobile elevated work platforms (MEWPs) increases the complexity of work at height. This study presents a UAV based monocular 3D approach for identifying unsafe operations from spatial relationships between workers and MEWPs. A scale aligned point cloud generation module reconstructs dense point clouds from video depth estimation and recovers global metric scale through calibration. A monocular 3D detector then predicts 7-DoF bounding boxes for workers and MEWPs. Unsafe operations are identified by geometry-based rules covering containment reduction by stance elevation and stability loss by center of mass shift. Field experiments on a construction site achieved centimeter level scale recovery, mean AP values of 0.38 and 0.52 under strict and loose 3D IoU thresholds, and 18.4 FPS. The results demonstrate the feasibility of metric consistent monocular UAV inspection without multi sensor setups.