改进CRNN的车牌识别方法-《集美大学学报（自然科学版）》

文章信息/Info

作者:: 林立雄1; 庄裕富2; 何洪钦3; 郑佳春1; 1．集美大学海洋信息工程学院，福建厦门 361021；2．福州大学机械工程及自动化学院，福建福州 350108；3．浙江大华技术股份有限公司，浙江杭州 310051

Author(s):: LIN Lixiong1; ZHUANG Yufu2; HE Hongqin3; ZHENG Jiachun1; 1．School of Ocean Information Engineerin，Jimei University，Xiamen 361021，China；2．School of Mechanical Engineering and Automation，Fuzhou University，Fuzhou 350108，China;3．Zhejiang Dahua Technology Co.Ltd.，Hangzhou 310051，China

Keywords:: license plate recognition; deep neural network; convolutional recurrent neural network(CRNN)

摘要:: 针对传统车牌识别方法在车牌模糊和光照复杂的场景下难以快速准确识别车牌信息的问题，为提高网络的特征提取能力，将带残差的ResNet引入卷积循环神经网络（convolutional recurrent neural network,CRNN)，提出了CRNN_ResNet车牌文本识别算法。该方法仅需0.012 s就能完成一张车牌图像的识别，在CCPD公开数据集上的识别准确率达到了98.5%。

Abstract:: Aiming at the problem that the traditional license plate recognition method was difficult to recognize license plate information quickly and accurately under the scene of fuzzy license plate and complex illumination,a fast license plate recognition method based on improved convolutional recurrent neural network（CRNN) was proposed.By introducing ResNet with residual into CRNN to improve the feature extraction ability of the network,this new method was named CRNN_ResNet license plate text recognition algorithm.This method only needed 0.012 s to recognize a license plate image,and the recognition accuracy reached 98.5% on the open data set of Chinese City Parking Dataset (CCPD).Keywords: license plate recognition；deep neural network；convolutional recurrent neural network(CRNN)