An Effective Method for Detecting Personal Protective Equipment at Real Construction Sites Using the Improved YOLOv5s with SIoU Loss Function

This study introduces a novel approach for detecting personal protective equipment (PPE) in real construction sites using YOLO backbones-Transformer. The proposed architecture is built from the YOLOv5s model by incorporating the SC3T module into the final layer for more accurate detection. Additionally, in the proposed YOLOv5s- Transformer model, the SIoU loss function is utilized instead of the CIoU loss function used in the original YOLOv5s model. Furthermore, we also introduce a new PPEVN dataset with 5 classes (person, helmet, vest, gloves, and boots) from images collected directly from real construction sites in Vietnam. Experimental results on our dataset show that the proposed model has Precision, recall, and mAP50 parameters of 0.848, 0.777, and 0.829, respectively, significantly improved over the original YOLOv5s model.