Artificial Intelligence-Based Application for Monitoring Safety Equipment in a Construction Sector Company

Abstract

This research focuses on the problem of occupational accidents due to the improper use of personal protective equipment (PPE) in the construction industry. The implementation of a detection model based on Artificial Intelligence is necessary to increase the efficiency and accuracy of monitoring these implements, as well as to reduce the incidence rate. During the development and training process of the model, 5,886 high-definition (HD) JPG images were collected in a dataset on the RoboFlow platform. For processing, these images were scaled to 640 x 640 pixels and were related to working environments in this sector, as well as different climatic conditions, spaces, and focuses. A comparison of the models was carried out. YOLOv8 presented a mAP of 93.38%, in contrast to SSD, which reached 34.22%, and Faster R-CNN, which achieved 41.99% on average during the first 50 training epochs. Subsequently, for the final review of the training, 400 epochs of the YOLOv8 model were completed, resulting in a mAP of 93.22%, a recall of 89.67%, and an accuracy of 91.18%. With tuning and training finalized, the model was used for the development of a web system, which was subsequently hosted on a cloud server to facilitate access. This tool promotes compliance with safety regulations during the execution of daily tasks in this sector