Traffic sign recognition in complex environments based on YOLOv5
Fei Feng, Yu Hu · 2023
To address the issue of difficulty in recognizing traffic signs in complex environments and the lack of accuracy, a traffic sign recognition model based on YOLOv5 in complex environments is proposed. Firstly, optimize the target detection layer and further reduce the size of the detection head to enhance the model's ability to extract features from small targets. Then, a cross-connection network is introduced into the network layer to effectively reduce the loss of feature information during network transmission. Finally, adaptive spatial feature fusion technology is used to strengthen the fusion of multi-scale feature information. Experimental results show that the traffic sign recognition in complex environments has reached 83.6%, 76.0%, and 82.2% in Precision, Recall and [email protected] respectively. Compared with YOLOv5s, these three indicators have increased by 6.4%, 3.1%, and 4.7% respectively. Moreover, the model's parameter amount has decreased by 5.6%, making the model more lightweight.