Enhancing construction safety management efficiency with AI-Powered real-time helmet detection

Abstract

To address the critical need for improved safety management in the construction industry, an AI-powered system for real-time safety helmet detection was developed in this study. A comprehensive dataset of 19,456 images was compiled and the YOLO object detection algorithm was employed to accurately identify workers who are not wearing helmets, thereby enabling prompt intervention and reducing the risk of head injuries on construction sites. The model's performance was further optimized through the application of transfer learning techniques, and rigorous evaluation procedures were conducted, which resulted in the achievement of 89% mAP, 89.6% precision, and 83.8% recall. This automated system is designed to improve safety management practices in the construction industry by automating the monitoring process, enabling real-time detection of non-compliance, and facilitating timely interventions. These features aim to reduce workplace accidents and promote a proactive approach to safety management. The study provides a practical tool for construction management professionals to enhance worker safety and support the adoption of preventive safety measures on construction sites.