Construcción de un sistema prototipo de monitoreo y alerta en tiempo real para el uso de implementos de seguridad industrial en centros de trabajo mediante visión artificial.

Abstract

This research project aimed to develop a real-time monitoring and alert prototype for use in industrial safety implementation in work centers using artificial vision. Prototype development consisted of 4 stages: training, feeding, data acquisition, and processing. In the first stage, the Yolov4-tiny convolutional neural network was trained using the pre-trained Darknet network to detect the different research classifications (helmet, vest, earmuffs, safety boots, gloves, and people) on the virtual Google Colab platform. In the second stage, the process of feeding the prototype was carried out using a 12v 3A adapter, which ensured the correct functioning of the prototype. In the third stage, data acquisition is carried out using the Raspberry Pi module Camera V2, 8Mpx, which was designed for better performance with the embedded system used in the fourth stage, which focuses on data processing using the 8GB Raspberry PI 4 Model B, which was programmed using Python with the detection algorithm based on Yolov4- tiny previously trained for detecting classifications in personal protective equipment. Once the different categories have been identified in the images obtained in the third stage, the prototype will discern if the personal protective equipment (PPE) used is correct or a correction needs to be made, resulting in images that will be stored both in the prototype as well as on Google Drive for further study of same. The results obtained through the prototype demonstrated the effectiveness of average precision per study classification as greater than 85%, which is significantly more effective than that shown by the pre-trained Yolo Coco Dataset, which has 57.9% effectiveness. In conclusion, the results obtained suggest that further modifications as part of future research should be considered.