Artificial intelligence applications in occupational health and safety

The construction industry has the highest number of injuries and fatalities in the workplace. In order to prevent fatal injuries and disabilities, it is important to implement advanced safety systems instead of traditional safety measures, as well as to check whether workers are using their protective equipment properly. The use of helmet, which is an effective measure against occupational accidents, also maintains its importance as a legal obligation. Classification and object detection by surveillance from videos or images through artificial intelligence-based computer vision offers a very common field of study. The mainstream focus on object detection recognizes that single-stage detectors outperform two-stage detectors in real-time estimation. In this study, the YOLOv9 model, which stands out with its speed and accuracy among single-stage detectors, was analyzed. In order to train the model, 3 different datasets were combined to create a single dataset consisting of 2107 images. The resulting dataset consists of images of helmets, people, vests, gloves and headphones reflecting real life in construction sites. Different versions of the YOLOv9 model were trained for 25, 50, 75, 100 epochs and the most successful performance was seen in the 100 epoch training of the gelan-e version. The results obtained were; precision 0,679, recall 0,772, F-1 score 0,72, mAP50 value 0,737, mAP50-95 value 0,431. In this study, the speed and accuracy performance of the YOLOv9 model was investigated for the detection of helmet use. The experimental results show that the model achieves a high performance in helmet use detection tasks, which has an effective role in preventing head injuries. The model also showed high performance not only in helmet detection, but also in the classification and detection of vests, earmuffs and unauthorized entry into the workplace.