Based on improved YOLOv11n and ST-GCN in complex occlusion video scenes Research on the Fall detection of the model

This paper proposes a fall detection method for industrial scenarios with complex occlusions. The improved YOLOv11-n (You Only Look Once V11-n) model achieves simultaneous human detection and skeletal keypoint localization through parallel processing. An optimized Spatial-Temporal Graph Convolutional Network (ST-GCN) is developed to capture fall patterns via spatiotemporal feature fusion. Experimental results on the LE2I Fall Detection Dataset demonstrate 94.8% accuracy with real-time performance at 32 FPS. The proposed framework effectively addresses multi-target recognition challenges in occluded environments, offering a practical solution for safety monitoring in industrial settings.