Workplace safety in the construction sector remains a critical concern, primarily due to low compliance with Personal Protective Equipment (PPE) standards. To address this, this study develops and evaluates a real-time PPE monitoring system, conducting a comparative analysis of two state-of-the-art object detection models: YOLOv8s and YOLOv11s. The system is designed to detect three essential PPE items: helmets, masks, and vests, and both models were trained on a custom dataset of 9,202 augmented images over 200 epochs. The final evaluation on an unseen test set revealed highly competitive performance. While YOLOv8s achieved a marginally higher mAP@0.5 (90.8%), YOLOv11s demonstrated superior precision (92.0%) and better performance on the stricter mAP@0.5:0.95 metric (54.4%). Based on this nuanced trade-off and its significantly higher computational efficiency (15% fewer parameters), YOLOv11s was selected as the optimal model. The chosen model achieved a real-time inference speed of approximately 112 FPS. A functional web-based prototype was developed using Flask to demonstrate the system's practical application. These findings confirm that YOLOv11s offers a more balanced and efficient solution for automating PPE compliance monitoring and highlight that a holistic evaluation beyond a single metric is crucial for deploying robust computer vision systems in real-world safety applications.
Akmal et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: