Advancing construction safety monitoring: video analysis of personal protective equipment using object detection and tracking approach

The construction industry ranks among the most hazardous, with high rates of workplace injuries and fatalities. Ensuring compliance with personal protective equipment (PPE) rules is critical for mitigating risks, yet compliance is often insufficient on construction sites. This research introduces a video-based proof-of-concept framework for detecting and tracking PPE, validated in a controlled laboratory environment designed to simulate various construction site scenarios using the YOLOv11 object detection model combined with a tracking algorithm. YOLOv11’s initial application on each video frame produces bounding boxes and confidence levels for PPE and personnel. The system then cross-references these detections across frames with similarity measures and appearance features. A distance-based spatial assignment method using Mahalanobis distance refines how PPE is allocated to individual workers. The framework was tested in various simulated conditions, including sunny, dark, low resolution, occlusion, and normal settings. Results from a 48-second period showed a reduction in false positives and standard deviation due to the tracking algorithm, with detection errors decreasing by 3 to 108 frames across most scenarios. YOLOv11 reached an average precision of 86.2% for PPE detection. Worker-specific allocation enabled detailed temporal analysis, demonstrating video-based tracking’s advantages over single-frame detection for proactive hazard identification.