Detecting personal protective equipment (PPE) is crucial for identifying, warning, and assessing safety on hazardous construction sites. This paper presents a computer vision-based approach to address this task. We have developed a comprehensive PPE dataset with nine classes, consisting of over 4000 images of workers wearing PPE at construction sites from various angles, distances, and positions. Our dataset includes not only traditional classes such as person, hardhat, vest, gloves, and boots but also critical classes like no hardhat, no vest, no gloves, and no boots, setting it apart from other datasets. We conducted experiments to evaluate and compare the performance of the latest YOLOv9 models. The results show that YOLOv9M achieved the highest mean average precision (mAP50) of around 80.8% for detecting PPE. In this study, we also designed a web interface with input options including images, videos, webcam, or YouTube links using the Streamlit library, providing PPE detection results as output.