Résumé : Safety clothing and helmets play a crucial role in ensuring worker safety at construction sites. Recently, deep learning methods have garnered significant attention in the field of computer vision for their potential to enhance safety and efficiency in various industries. However, limited availability of high-quality datasets has hindered the development of deep learning methods for safety clothing and helmet detection. In this work, we present a large, comprehensive, and realistic high-quality dataset for safety clothing and helmet detection, which was collected from a real-world chemical plant and annotated by professional security inspectors. Our dataset has been compared with several existing open-source datasets, and its effectiveness has been verified applying some classic object detection methods. The results demonstrate that our dataset is more complete and performs better in real-world settings. Furthermore, we have released our deployment code to the public to encourage the adoption of our dataset and improve worker safety. We hope that our efforts will promote the convergence of academic research and industry, ultimately contribute to the betterment of society.