An Investigation of the Use of Personal Protective Equipment on the Jobsite using Transfer Learning Approach

Abstract

Practicing engineers throughout the globe are very concerned about the safety at the construction sites. There are still dangers in construction workplaces and thus requires suitable detection and risk assessment. This research establishes a framework for real-time monitoring of Personal Protective Equipment (PPE) compliance by construction workers, which is meant to be incorporated into an organization's safety workflow. Using the Convolutional Neural Network (CNN) model which has been constructed using multiple transfer learning approaches, this inquiry study takes the existence of hard helmets and safety feature jackets as a part of its analysis. Four types of compliances are predicted by the model, including; 1. Red Sign (NOT SAFE) 2. Green Sign (SAFE) 3. Yellow Sign (No Hard Hat or Gloves and something missing) 4. Dark Yellow (No Jacket and many protectives missing). Overall, the research shows that computer vision-based approaches through transfer learning may be used to automate safety-related compliance procedures at construction sites. It is also clear that real-time performance is better, with a higher accuracy rate after review and study of the obtained consequences by different transfer learning approaches.