采用视觉显著性和卷积网络的车牌定位算法

A License Plate Location Algorithm Using Visual Saliency and Convolutional Network

针对传统的车牌定位方法存在对环境要求高,容易受干扰,鲁棒性不强等不足,提出将视觉显著性与卷积神经网络相结合应用于车牌定位,设计了一种车牌定位方法。该方法包含2个阶段,第1阶段为提取车牌候选区域,将视觉注意机制引入车牌定位过程,采用自下而上的模型,提取出车辆图像中具有显著特征的车牌区域。采用视觉显著性算法,实现车牌候选区域的快速定位和提取,避免了在车辆图像的各个区域进行扫描。第2阶段为车牌识别,将深度卷积神经网络应用于车牌区域识别,实现候选区域中非车牌和车牌的准确分类。实验结果表明,该方法性能优异,大幅降低了车牌区域漏检率,鲁棒性强,定位准确率比数字图像方法提高约5%。

In view of the shortcomings of traditional license plate location methods,such as high environmental requirements,easy interference and poor robustness,we design a license plate location method by combining visual saliency and convolutional neural network. The method consists of two stages. The first stage is to extract the license plate candidate area. The visual attention mechanism is introduced into the license plate location process,and the bottom-up model is used to extract the license plate area with significant features in the vehicle image. The visual saliency algorithm is used to realize the rapid positioning and extraction of the license plate candidate area,thus avoiding scanning in various areas of the vehicle image. The second stage is license plate recognition. The deep convolutional neural network is applied to the identification of the license plate area for accurate classification of non-license plates and license plates in the candidate area. The experiment shows that the proposed method has excellent performance,sharply dropping the missed detection rate of the license plate area,with strong robustness,and its accuracy of positioning is improved about 5% compared with the digital image method.