RT-LPDRnet: A Real-Time License Plate Detection and Recognition Network

Abstract

Benefiting from the rapid development of deep learning, the accuracy of license plate (LP) detection and recognition has been greatly developed. However, the low computing power of embedded devices is difficult to complete this task in real-time. To tackle this problem, we propose a real-time license plate detection and recognition network in this paper. The detection head we designed can detect the bounding box and four corner points of the license plate, and then we apply the ROIAlign method to extract the features on the same backbone in order to perform license plate recognition. The resulting architecture, called RT-LPDRnet, outperforms all the SOTA methods on the large-scale license plate data set Chinese City Parking Dataset (CCPD), meanwhile with faster inference time than recent methods. Our code will be publicly available.