Ferramenta baseada em visão computacional para monitoramento automático do uso de EPIS em ambiente industrial

Abstract

Occupational safety in industrial environments, especially in the mineral water bottling sector, requires compliance with sanitary regulations and proper use of Personal Protective Equipment (PPE). In critical production areas, the absence of such equipment may compromise worker safety and process compliance. This work developed and evaluated a tool for automatic PPE recognition using computer vision and deep learning. The tool was based on the YOLOv8n detection model, trained with representative images from the industrial environment and controlled scenarios. The implemented architecture separates processing and actuation. Images are analyzed on a dedicated computer through a Python application with a Tkinter interface, while the embedded module, based on the Raspberry Pi Pico W, receives states through web requests and activates a colored light stick. The detection model performance and the integrated operation of the system were evaluated. YOLOv8n achieved an mAP@0.5 of 0.911, and the average time between image capture and physical response was approximately 140 ms. The results indicate technical feasibility to support automatic PPE verification in an industrial environment, considering limitations related to lighting, occlusions, camera distance, and training data diversity.