Human fall detection in industrial environment using YOLOv۸ with transfer learning

Ensuring worker safety in industrial environments, particularly in high-risk areas like pipe production lines, necessitates reliable fall detection systems. This research introduces a novel two-step model employing the YOLOv۸ framework for real-time fall detection. Initially, a YOLOv۸ model is trained to detect the presence of persons within the production line using annotated video data. Subsequently, a second YOLOv۸ model is trained to classify these detected instances into 'fall' and 'no fall' categories. To optimize real-time performance, a frame-skipping technique is integrated, reducing computational demands while preserving detection accuracy. Furthermore, the system employs a Region of Interest (ROI) approach, focusing on specific hazardous zones where falls are more likely to occur, thereby enhancing the model's efficiency and relevance. Experimental results indicate that the proposed method achieves high precision and recall and F۱ in fall detection, demonstrating its robustness and effectiveness.

Keywords:

component , Fall Detection , YOLOv۸ , Industrial Safety , Real-Time Monitoring , Object detection