AI-Based Personal Protective Equipment Compliance Detection and Access Control Using Deep Learning

This paper presents an AI-based Personal Protective Equipment (PPE) compliance detection and access-control system designed for industrial and construction environments. The proposed system uses a trained Ultralytics YOLO object-detection model to identify workers and PPE items from images or camera frames, followed by a person-level PPE decision engine that determines whether each worker satisfies the required safety conditions. The access-control decision is based specifically on three mandatory PPE items: safety helmet, safety vest, and safety boots. A worker is classified as ACCESS ALLOWED only when all three required items are confidently detected; otherwise, the worker is classified as ACCESS DENIED and the missing PPE items are reported. Gloves and goggles may be detected by the object-detection model but do not affect the access decision. The system was implemented using Python, PyTorch, CUDA, Ultralytics YOLO, OpenCV, and Streamlit, and was evaluated using a construction PPE dataset containing training, validation, and test images. Experimental testing demonstrates the feasibility of combining object detection with rule-based person-PPE association for automated safety compliance monitoring and access control. The proposed approach provides a safety-oriented framework that can be extended for real-time camera-based monitoring in industrial and construction environments.