YOLOv10 for Real-Time Detection of Personal Protective Equipment on Construction Workers

This study addresses the challenges of detecting Personal Protective Equipment (PPE) on construction sites, where work-related accidents frequently occur due to the incomplete use of PPE, which can lead to fatal outcomes. The objective of this research is to evaluate the use of the YOLOv10 model a lightweight and efficient object detection architecture to detect various PPE items: safety helmets, safety vests, gloves, and safety boots. The dataset consists of 1,620 images and was split using two configurations: 70:20:10 and 80:10:10 for training, validation, and testing sets, respectively. The YOLOv10 model was evaluated using the key metric of Mean Average Precision (mAP). The evaluation results demonstrate the model’s capability to accurately detect PPE, despite variations in data splitting and the number of epochs used. The findings show that the YOLOv10 algorithm performs very well in detecting PPE. On manually processed datasets, the YOLOv10-M model achieved a mAP50 of 97.8% with a 70:20:10 split and 98.4% with an 80:10:10 split. Meanwhile, on datasets processed using Roboflow, the YOLOv10-B model obtained a mAP50 of 85.2% with the 70:20:10 split, and the YOLOv10-S model reached 84.6% on the 80:10:10 split. These findings indicate that YOLOv10 delivers a significant performance improvement in PPE detection compared to previous approaches. The algorithm’s ability to achieve high mAP50 scores under certain conditions highlights its potential for real-time implementation in construction environments, where accurate and timely PPE detection is crucial to reducing future workplace accidents