Implementación de un sistema basado en visión artificial para la verificación del uso correcto de equipos de protección personal y el monitoreo de acceso a zonas de riesgos en entornos industriales

Abstract

This thesis presents the development and implementation of a computer vision–based system for the automatic detection of Personal Protective Equipment (PPE) using deep learning techniques. The main objective is to enhance workplace safety by identifying, in real time, the correct use of protective elements such as safety helmets and reflective vests in industrial environments. A dedicated dataset was created and manually labeled to train an object detection model based on the YOLO11s architecture. This model was selected due to its balance between inference speed and detection accuracy, making it suitable for low-resource hardware platforms while maintaining near real-time performance. The training process was conducted using a structured configuration defined in a data.yaml file, including training, validation, and test data splits. Model performance was evaluated using standard metrics such as Precision, Recall, mAP@0.5, and confusion matrix analysis. The results demonstrate high detection capability, with strong precision and recall values, indicating stable convergence and reliable generalization on validation data. The proposed system proves that deep learning–based object detection models can be effectively integrated into cost-efficient safety monitoring solutions, contributing to accident prevention and the reinforcement of occupational safety culture in industrial settings.