基于 YOLO 的无约束场景中文车牌检测与识别.

Chinese License Plate Detection and Recognition in Unconstrained Scenarios Based on YOLO.

In view of the problems of traditional Chinese license plate recognition methods, such as the requirement of scenes, poor real-time performance, and inability to deploy on edge devices, this study proposes a Chinese license plate detection and recognition method based on YOLO(You Only Look Once) in unconstrained scenes. This method is divided into two modules: license plate detection and license plate character recognition. In the license plate detection part, the improved YOLOv5 model is used to predict four groups of key points for license plate correction based on the prediction of target candidate regions, and the pre-training model trained on the COCO data set is used for training, which reduces the error detection problem caused by the complex environment and has high real-time performance. In the license plate character recognition part, the CRNN(Convolutional Recurrent Neural Network) model is improved, which greatly reduces the parameters and computation of the algorithm, so that it can be successfully deployed in various edge devices. Experimental results show that the proposed method can efficiently detect and recognize license plates in complex environments. The map value of the proposed license plate detection model is 3.0% higher than that of Retina-face in the license plate detection data set. Compared with LPR-Net, the accuracy of license plate character recognition model in license plate recognition data set is improved by 4.2%. [ABSTRACT FROM AUTHOR]

传统中文车牌识别方法对场景约束有要求, 算法实时性差, 且无法被部署在边缘设备上。针对上述问题, 文中提出一种基于YOLO(You Only Look Once)的无约束场景中文车牌检测与识别方法。该方法分为车牌检测和车牌字符识别两个模块。在车牌检测部分, 使用改进的YOLOv5模型, 在预测目标候选区域的基础上多预测4组关键点用于车牌矫正, 并使用在COCO数据集上训练的预训练模型进行训练, 减少了由环境复杂引起的误检问题, 具有高实时性。在车牌字符识别部分, 改进了CRNN(Convolutional Recurrent Neural Network)模型, 减少了算法的参数量和计算量, 使其能成功部署于各类边缘设备。实验结果表明所提出的车牌识别方法能在复杂环境中高效检测并识别车牌。文中提出的车牌检测模型在车牌检测数据集上的map值相较Retina-face提升了3.0%, 车牌字符识别模型在车牌识别数据集上精确度相比LPR-Net提升了4.2%。 [ABSTRACT FROM AUTHOR]

Copyright of Electronic Science & Technology is the property of Electronic Science & Technology Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)