Sistem Deep-Learning Yolov8 untuk Deteksi Penggunaan APD Secara Real-Time

Although workplace safety regulations in construction are clear, many workers are still reluctant to use Personal Protective Equipment (PPE) due to a lack of awareness, work pressure, and limited facilities. As a result, the risk of serious accidents increases. Conventional approaches such as verbal warnings or CCTV monitoring are considered less effective for early detection and prevention of violations. This study proposes an automatic detection system for PPE usage in construction areas using YOLOv8. The model was trained on a secondary dataset of 3,569 images for 100 epochs, with a 60% training, 20% validation, and 20% test split. Testing on 90 real-time frames showed good performance in detecting 8 PPE classes, with an average precision of 0.935, recall of 0.806, and F1-measure of 0.862. The results indicate that the system can classify PPE usage with high accuracy. However, a recall below 1 suggests that some objects, particularly "not wearing glasses" and "not wearing shoes," failed to be detected. The F1-measure of 0.862 reflects a good balance between precision and recall.