The construction industry is inherently high risk, with worker safety being a paramount concern. This project focuses on developing a real-time safety monitoring system for construction sites using vision-based techniques to enhance worker safety. The methodology begins by collecting diverse images and annotating each image with relevant safety elements. In the data pre-processing stage, image augmentation techniques such as rotation, scaling, and flipping enhance dataset diversity, improving the model’s robustness. Following augmentation, the images are normalized to standardize inputs for the YOLO algorithm. The dataset is then divided into training, validation, and test sets, ensuring a balanced distribution of scenarios for effective training and evaluation, ultimately facilitating accurate real-time monitoring. The model is trained on the annotated dataset. The mean Average Precision (mAP), Precision, and Recall are evaluated. The ultimate goal is to develop a system capable of accurately detecting safety risks, reducing accidents, and ensuring compliance with construction site safety standards.