A License Plate Detection and Recognition Method Based on Improved Yolov5

Abstract

This study proposes a license plate detection and recognition method based on yolov5. In order to accurately detect the license plate and accurately identify the license plate number, this study adds key point detection based on the yolov5 algorithm to correct the license plate image into a rectangle through perspective transformation, which greatly improves the accuracy of license plate number recognition. Due to embedded deployment, this study simplifies the yolov5 network structure to a certain extent. While reducing the network depth and width, ShuffleNet Block is used to replace the C3 module in yolov5. In the end, this study achieved a mean average precision (mAP) of 89.7% while controlling the network calculation amount to 5GFLOPs. Therefore, this model has broad application prospects in the transportation field.