AI-Powered Real-Time Monitoring System for Health Care Personal Protective Equipment (PPE) Compliance

Abstract

Adhering to Personal Protective Equipment (PPE) protocols remains one of the most difficult aspects of healthcare, as manual supervision is time-consuming, error-prone, and subjective. This study aims to overcome the above challenge by creating an AI-based real-time computer vision system for automated PPE compliance monitoring and instant notification of non-compliance. Using computer vision and deep learning frameworks, a lightweight YOLOv5n model was trained to detect 6 PPE classes and their non-compliance cases: lab coat, nose mask, and gloves, non-compliant of lab coat, nose mask, and gloves. An experimental research design was used. The study population comprised image data, and samples were annotated from these images with standard labelling procedures. Precision, recall, F1 score, and mean Average Precision (mAP) were used as metrics for model validation. The whole system was developed on a Raspberry Pi with an RTSP camera and a buzzer alert system. Results show that the real-time performance is satisfactory on low-power hardware with class accuracy of 88% (lab coat), 87% (nose mask), 65% (gloves), 87% (no nose mask), 81% (no gloves), and 68% (no lab coat). The results show that automated monitoring of PPE can increase compliance, reduce the need for human observers, and improve safety outcomes. The study wraps up by suggesting model optimization, multi-person tracking, and remote alert integration as ways to expand deployment.