Application and validation of RT-DETR target detection model for site safety management in real-time helmet wearing monitoring system

Abstract

Safety helmet wearing detection using video real-time monitoring system is important for site safety. The existing safety helmet wearing detection algorithms have more application scenario condition limitations, and it is difficult to meet different scenario requirements at the same time. In order to solve the above problems, the article proposes an improved safety helmet wearing target detection model based on RT-DETR model. The model reconstructs the original backbone network BackBone with ConvNeXt, introduces the F-CBAM attention mechanism to improve the feature extraction effect of the model on small targets and low-resolution images, and improves the original GIoU loss function to CIoU loss function, so as to improve the convergence efficiency and accuracy of the model. The results show that the size of the improved RT-DETR model is only increased by 1.41M compared with the YOLOv7 model, the mAP reaches 93.04%, and the inference time of the model is only 34 frames/ms.Relying on the improved RT-DETR model, the detection effect of helmet wearing can be significantly improved, and the overall generated model is smaller, which can satisfy the practical deployment in different scenarios.