复杂环境下的车牌识别算法研究  

Research on License Plate Recognition Algorithm under Complex Environments

作　　者：陈民 吴观茂[1] CHEN Min;WU Guanmao(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan Anhui 232001)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《湖北理工学院学报》2023年第4期52-56,共5页Journal of Hubei Polytechnic University

基　　金：安徽省自然科学基金项目(项目编号:1908085MF189)。

摘　　要：为解决复杂环境下车牌检测与识别模型处理速度慢、检测精度低和模型参数过大的问题,提出了一种端到端的车牌识别算法。首先采用MSR图像增强方法对车牌图像进行预处理,然后使用最新的YOLO系列算法YOLOX,以35帧/s的速度检测出图像目标区域,使用改进的Bi-GRU-CRNN识别模型进行车牌字符识别。在国内车牌识别开源数据集中随机选取5000张图片进行实验,发现相较于其他车牌识别模型,改进后的模型识别速度更快,识别准确率达到98%以上。An end-to-end license plate recognition algorithm is proposed to solve the problems of slow processing speed,low detection accuracy and oversized model parameters of license plate detection and recognition models under complex environments.Firstly,MSR image enhancement method is used to pre-process the license plate images,and then the latest YOLO series algorithm YOLOX is used to detect the image target region at a speed of 35 frames/s,and the improved Bi-GRU-CRNN recognition model is used for license plate character recognition.5000 pieces of images were randomly selected from the Chinese City Parking Dataset for experiments,finding that the improved model was faster and the recognition accuracy reached over 98%compared to other license plate recognition models.

关 键 词：YOLOX算法 卷积神经网络 车牌识别 Bi-GRU识别模型 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]