Detection of Traffic Signs in Complex Weather Conditions Based on YOLOv5
Tong Jiang, Yanhua Xian · 2023
Accurate and real-time detection of traffic signs in real-world scenarios is of paramount importance for traffic safety. To address the low precision issue when detecting small target traffic signs using YOLOv5 algorithm, this study proposes a detection algorithm specifically designed for small target traffic sign detection. Firstly, an SE module is introduced into the main network to focus on learning critical information within features. Secondly, a CoT module is incorporated into Neck network to facilitate comprehensive feature extraction of contextual information. Finally, a small target detection layer is introduced to enhance the model’s ability to detect small targets. Experimental results on TT100K dataset demonstrate that the proposed improved model achieves a 1.9% increase in accuracy, reaching 86.1%; a 2.3% increase in recall, reaching 79.5%; a 2.4% increase in mAP, reaching 85.5%; and an FPS of 33.4, meeting real-time requirements.