Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (2): 240-252.DOI: 10.3778/j.issn.1002-8331.2108-0439

• Graphics and Image Processing • Previous Articles     Next Articles

License Plate Detection Using Siamese Feature Pyramid and Cascaded Positioning

ZHANG Junqing, XIONG Yujie, SUN Xiankun, GAO Yongbin   

  1. 1.School of Electric and Electronic Engineering, Shanghai University of Engineering Science, Shanghai 201620, China
    2.Shanghai Key Laboratory of Multidimensional Information Processing, East China Normal University, Shanghai 200241, China
  • Online:2023-01-15 Published:2023-01-15



  1. 1.上海工程技术大学 电子电气工程学院,上海 201620
    2.华东师范大学 上海多维度信息处理重点实验室,上海 200241

Abstract: In order to solve the problem of license plate detection in complex environment, a novel license plate detection algorithm using siamese feature pyramid and cascaded positioning is proposed. Firstly, the original features of the images are extracted by the residual network with the Mish activation function. At the same time, a siamese feature pyramid network(SFPN) is proposed in this paper. The extracted features are sent to the siamese feature pyramid network for multi-level feature fusion. Then, a region proposal network based on the prior of shape is used to generate the regions of interest. Finally, the regions of interest are fed to the cascaded positioning network to obtain the license plate location. The experimental results show that the proposed algorithm effectively improves the performance of license plate detection on both AOLP and CCPD datasets.

Key words: license plate detection, deep learning, siamese feature pyramid, cascaded positioning

摘要: 为了解决复杂环境中不同因素干扰车牌检测精确度的问题,提出了一种基于双金字塔特征融合的复杂环境下车牌检测算法。通过采用Mish激活函数的残差网络(ResNet101-M)对输入图像进行初级特征提取;在传统特征金字塔网络(feature pyramid network,FPN)的基础上,提出了一种改进的双金字塔特征融合网络(siamese feature pyramid network,SFPN)。被提取的初级特征被送入该网络进行多层特征融合。融合后的特征被送入基于形状先验的锚点设置网络来确定感兴趣区域。将所生成的感兴趣区域送入级联定位网络从而得到准确的车牌检测结果。实验结果表明,该算法在AOLP与CCPD车牌数据集上均能够有效提升检测性能。

关键词: 车牌检测, 深度学习, 双金字塔特征融合, 级联定位