Automatic Detection of Personal Protective Equipment in Industrial Environments Using YOLO and the NAO Humanoid Robot

Resumo

Occupational safety in industrial environments represents a critical challenge that requires robust and efficient technological solutions. This work presents the development and implementation of a proof-of-concept system for detecting Personal Protective Equipment (PPE) using the You Only Look Once (YOLO) algorithm integrated with the NAO humanoid robot. The model was trained via fine-tuning on a specialized dataset containing annotations for different PPE classes, including safety helmets, reflective vests, safety boots, gloves, and masks. System validation was performed on five synthetic images generated by Artificial Intelligence under controlled conditions. Model-level inference times ranged from 0.01 to 0.04 seconds, while the implemented application executed detection cycles every 2 seconds. The obtained results show confidence rates exceeding 85% for the main PPE classes, notably for Safety Vest (90.05%), Person (87.93%), and Hardhat (86.17%). Integration with the NAO robot demonstrates the feasibility of combining robotic visual sensing with offboard deep-learning inference for occupational-safety monitoring. The study therefore establishes a proof-of-concept basis for future validation in real industrial environments. The analysis focused on three main dimensions: class confidence, processing latency, and detection coverage across controlled synthetic scenes.