Computer Vision-Based PPE Compliance Detection in a Pharmaceutical Laboratory

The use of Personal Protective Equipment (PPE) is important for maintaining practitioner safety in pharmaceutical laboratories; however, manual inspection may lead to oversight. This study aims to develop a system capable of automatically detecting the use of masks, laboratory coats, and gloves and providing detection results. The system was developed using a computer vision approach with the YOLOv8s model implemented on the NVIDIA Jetson Orin NX and trained using five classes: mask, labcoat, glove, no_glove, and no_labcoat. The detection results were used to determine PPE completeness, provide notifications through an LCD and speaker, and record violations through the Firebase Realtime Database for subsequent display on a Progressive Web Application (PWA). Evaluation results show that the model achieved 89% accuracy and 96.5% mAP@0.5, while testing on 40 samples resulted in a 95% detection success rate. The system also operated continuously for 8 hours without lag or crashes, with an FPS of 14–16 and power consumption of 5.7–9.2 W. These results indicate that the developed system can perform real-time PPE detection and monitoring with stable performance and can support PPE compliance supervision in a pharmaceutical laboratory.