Vision-enabled detection of safety helmet compliance in construction zones

In the rapidly evolving field of construction management, worker safety remains a top priority. This paper introduces an innovative vision-based system for real-time detection of helmet compliance, specifically designed for construction sites, utilizing advanced computer vision techniques and machine learning algorithms within the YOLO (you only look once) framework. Our system leverages high-resolution video feeds from strategically positioned cameras to monitor adherence to safety regulations regarding helmet usage. By employing deep learning methodologies, the system effectively identifies individuals not wearing helmets, thereby significantly mitigating the risk of head injuries among workers. Our training and validation results revealed an impressive precision exceeding 97% at mAP@0.5 for both helmeted and non-helmeted individuals. Furthermore, our experiments demonstrate exceptional detection accuracy, demonstrating the system's resilience under varying lighting conditions and diverse worker movements. The consistent decrease in loss and improvement in metrics throughout training validates the effectiveness of the YOLOv8 model in enhancing recognition performance. The implications of this research extend beyond mere regulatory compliance, opening avenues for innovative applications in occupational safety management. This study highlights the critical role of technology in protecting lives and lays the groundwork for future advancements in smart construction environments.