Personal Protective Equipment Detection for Construction Site Safety Using YOLOv9

Worker safety in construction sites hinges on effective Personal Protective Equipment (PPE) use, yet ensuring consistent adherence remains challenging. This research underscores YOLOv9's potential to automate PPE detection, crucial for enhancing safety compliance and mitigating construction site hazards effectively. This study investigates a PPE detection system employing the YOLOv9 object detection algorithm to bolster safety protocols. Evaluating two variants-compact (YOLOv9-c) and extended (YOLOv9-e)-using the Roboflow dataset, results indicate YOLOv9-c achieved 0.934 precision, 0.85 recall, and 0.905 mAP50 with SGD, while Adam showed higher recall and mAP50 but slightly lower precision. YOLOv9-E with Adam excelled with 0.925 precision, 0.874 recall, and 0.923 mAP50, outperforming SGD. Challenges include data imbalance and static testing set limitations, urging future work on dataset balance, optimizer exploration, and robustness testing for real-world applications.