Improved Traffic Sign Detection Algorithm for YOLOv5s

Traffic sign detection is an integral part of advanced assisted driving systems and intelligent transportation systems. Based on the TT100K dataset, this paper conducts a study on the improved YOLOv5s traffic sign detection algorithm. Firstly, to address the problem of low detection accuracy due to unbalanced sample size among different categories in the TT100K dataset, the replacement method, the paste method and the conventional data expansion method were successively used on the dataset, resulting in an increase in traffic sign detection accuracy from 80.9% to 91.2%. In order to further improve the application value of the traffic sign detection algorithm, this paper reduces the downsampling layer of the backbone in YOLOv5s to obtain a detection head with higher resolution, thus getting the model YOLOv5s-strong, whose detection accuracy is 4.9% higher than YOLOv5s. Even surpasses the detection accuracy and inference time of YOLOv7 with less number of parameters and calculations