Multi-Class Personal Protective Equipment Detection for Construction Safety Using YOLOv11

Construction sites remain hazardous environments where inadequate personal protective equipment (PPE) compliance contributes significantly to injuries and fatalities. This study presents a YOLOv11-based system for automated multi-class PPE detection at construction site entry points. The system detects six categories: safety helmets, vests, shoes, and their non-compliant alternatives. We trained YOLOv11n on 352 annotated images using 7:2:1 split with COCO pretrained weights. The system achieves overall mAP50 of 0.602, with performance variation across classes—safety vests reach 0.834 mAP50 while footwear proves most challenging at 0.487 mAP50. The integrated system provides real-time detection at 25-30 FPS, visualization dashboards, and automated alerts. Results demonstrate proof-of-concept viability, with dataset scale identified as the primary constraint. This work contributes to proactive construction safety through continuous, automated PPE compliance verification.