Two Novel Models for Traffic Sign Detection Based on YOLOv5s

Object detection and image recognition are some of the most significant and challenging  branches in the field of computer vision. The prosperous development of unmanned driving technology has made the detection and recognition of traffic signs crucial. Affected by diverse factors  such as light, the presence of small objects, and complicated backgrounds, the results of traditional  traffic sign detection technology are not satisfactory. To solve this problem, this paper proposes two  novel traffic sign detection models, called YOLOv5-DH and YOLOv5-TDHSA, based on the  YOLOv5s model with the following improvements (YOLOv5-DH uses only the second improvement): (1) replacing the last layer of the ‘Conv + Batch Normalization + SiLU’ (CBS) structure in the  YOLOv5s backbone with a transformer self-attention module (T in the YOLOv5-TDHSA’s name),  and also adding a similar module to the last layer of its neck, so that the image information can be  used more comprehensively, (2) replacing the YOLOv5s coupled head with a decoupled head (DH in both models’ names) so as to increase the detection accuracy and speed up the convergence, and  (3) adding a small-object detection layer (S in the YOLOv5-TDHSA’s name) and an adaptive anchor  (A in the YOLOv5-TDHSA’s name) to the YOLOv5s neck to improve the detection of small objects.  Based on experiments conducted on two public datasets, it is demonstrated that both proposed  models perform better than the original YOLOv5s model and three other state-of-the-art models  (Faster R-CNN, YOLOv4-Tiny, and YOLOv5n) in terms of the mean accuracy (mAP) and F1 score,  achieving mAP values of 77.9% and 83.4% and F1 score values of 0.767 and 0.811 on the TT100K  dataset, and mAP values of 68.1% and 69.8% and F1 score values of 0.71 and 0.72 on the CCTSDB2021  dataset, respectively, for YOLOv5-DH and YOLOv5-TDHSA. This was achieved, however, at the  expense of both proposed models having a bigger size, greater number of parameters, and slower  processing speed than YOLOv5s, YOLOv4-Tiny and YOLOv5n, surpassing only Faster R-CNN in this  regard. The results also confirmed that the incorporation of the T and SA improvements into YOLOv5s  leads to further enhancement, represented by the YOLOv5-TDHSA model, which is superior to the  other proposed model, YOLOv5-DH, which avails of only one YOLOv5s improvement (i.e., DH).