Find in Library

Traffic sign detection is a crucial component of the autonomous driving field, where real-time performance and accuracy play a significant role in ensuring vehicle safety. This paper aims to improve the detection performance of multi-scale traffic sign targets and proposes an enhanced multi-scale traffic sign detection algorithm MMW-YOLOv5 based on the YOLOv5 algorithm. The algorithm first uses a multi-scale fusion network (MSFNet) on the neck, which significantly enhances the algorithm’s fusion capabilities for multi-scale features and its ability to detect small-sized targets. Secondly, the C3 bottleneck structure in the trunk and neck used to process small-scale feature maps is replaced with the multi-scale feature extraction bottleneck module (MSFEBM) to obtain rich multi-scale feature information and facilitate multi-scale target detection. Finally, the positioning regression function Wise-MPDIoU (WMPDIoU) is used to further improve the overall accuracy of the model and accelerate the convergence speed of the network. Experimental results show that the detection accuracy of the MMW-YOLOv5 algorithm on the TT100K data set reached 87.1% mAP@0.5 and 53.7% mAP@0.5:0.95, which were improved by 6.6% and 5.1% respectively compared with the baseline model.