Research On the Application of Deep Learning-Based License Plate Computer Recognition System in Traffic Scenarios

Abstract

As the development of machine vision technology matures, this paper builds a set of license plate recognition algorithms based on deep learning algorithms and applies them in traffic monitoring scenarios, which can accurately identify the license plate information of passing vehicles at traffic intersections, and the detection accuracy mAP can reach more than 99%, The recognition accuracy can reach more than 94%, the detection algorithm uses YOLOV4, which can accurately detect the position information in the image where the license plate is located, the recognition algorithm uses CRNN, the detected ROI area is cut out for text recognition, and the algorithm is deployed and used Deployed with the tensorrt acceleration library and deployed end-to-end on the NVIDIA Jetson AGX Xavier embedded edge computing device. It only takes 17ms to identify all the license plate number information in an image.