Safety Helmet Detection of Workers in Construction Site using YOLOv8

The safety of construction workers is a paramount concern in the modern construction industry. A significant proportion of injuries and fatalities on construction sites are attributed to a lack of adherence to safety regulations, often resulting from the non-utilization of Personal Protective Equipment (PPE) by workers. Additionally, effective monitoring of construction areas by site supervisors can be challenging due to the vastness and complexity of construction sites. Manual monitoring, while a crucial aspect of construction safety management, is often hindered by time constraints and associated costs. To address these challenges, the application of computer vision and Convolutional Neural Network(CNN) techniques has led to the development of automated helmet and jacket detection systems. This research employs the YOLOv8 object detection algorithm to explore the efficacy of safety helmet detection, aiming to enhance the accuracy of automated safety helmet detection systems. The YOLOv8 object detection algorithm is a robust tool for recognizing objects in images. Our dataset comprises a total of 4,200 images, including 4,000 images of hard hats and jackets obtained from Kaggle [25], supplemented by 200 images depicting foggy, rainy, and nighttime conditions from our custom collection sourced from the internet. The dataset was meticulously divided into training, testing, and validation sets, following a ratio of 60%, 20%, and 20%, respectively. The experimental results revealed that the YOLOv8m architecture achieved an accuracy of 92%, demonstrating its effectiveness in detecting safety helmets under various lighting conditions, including low-light environments. This performance highlights the potential of the proposed approach for real-world applications in construction site safety monitoring. This developed model primarily focuses on detecting helmets and jackets in real-time in adverse weather conditions such as rainy, foggy, and low-light environments. It triggers an alarm if any workers are not wearing their helmets and jackets. We have also discussed model training, system structure, and performance measures.