Multi-scale Traffic Sign Detection and Recognition Method Fused with Attention Mechanism

For the problems of the traditional methods in the process of small target detection of traffic signs, such as being susceptible to environmental factors, poor real-time performance and weak generalization ability, this paper proposes an improved road traffic sign detection and recognition method based on YOLOv5 target detection model. Firstly, the original target box in YOLOv5 network which was not suitable for this detection task is improved, and the clustering method of target box obtained from the original network was optimized to K-Means++ clustering method to generate new anchor coordinates. Secondly, the CSP structure of YOLOv5 network was replaced by the CSPGhost structure composed of Ghost module. The CBAM attention mechanism module is added behind the backbone network to improve the image feature extraction ability of the network model. Finally, the detection layer structure of the model is improved, and the detection scale of the target is increased to improve the detection accuracy of the model for small targets. In order to verify the effectiveness of the improved network, a model comparison experiment is designed in this paper. It shows that 45 types of traffic signs are detected on TT100K traffic sign dataset. The experimental results show that the accuracy of the improved network is improved by 7.26% compared with YOLOv5 network.