YOLO based One-stage license plate detection

In recent years, traffic facilities have been gradually improved, and the types and numbers of vehicles have been increasing, which has resulted in higher requirements for abnormal vehicle monitoring and vehicle parking scheduling. In this paper, we use a One-stage target detection algorithm to recognize license plates and use YOLOv5 based method for car license plate detection. We optimized the model by adding a multiscale attention mechanism based on the idea of exponential moving average to the neck. We used the improved YOLOv5, which was tested on the CCPD dataset, and the precision reached up to $\mathbf{9 9. 9 2 \%}$. In the detection for images with $640 \times 640$ resolution, the detection time of the model can reach 0.026s/frame.