A real-time system for detecting personal protective equipment compliance using deep learning model YOLOv5

Abstract

The manufacturing and construction sectors report high rates of occupational accidents and diseases. In response, many countries, including Malaysia, have incorporated the use of Personal Protective Equipment (PPE) into their government laws and safety regulations. In Malaysia, this requirement is outlined in Standard Operating Procedure (SOP) manuals in the construction sector, which specify the mandatory use of seven types of PPE: hard hats, safety goggles, high-visibility vests, safety shoes, gloves, respirators, and earplugs. However, monitoring a large number of workers and ensuring consistent use of PPE on construction sites can be a challenging task if done manually by site supervisors. Therefore, this project aims to implement a real-time PPE detection system using a deep learning model to effectively monitor workers’ safety compliance. This system can generate reports that companies in the appropriate sectors can use to monitor compliance with the Occupational Safety and Health Act (OSHA). The prototype system was developed as a web-based platform. The object detection deep learning model achieved a mean Average Precision (mAP) of 91.5% in the metric evaluation, while the accuracy testing yielded a 98.3% accuracy rate. For future work, it is suggested to enhance the detection model for other types of PPE and to make the system more reliable across all sectors by expanding the training dataset.