License plate detection is a key problem in intelligent transportation systems. Recently, many deep learning-based networks have been proposed and achieved incredible success in general object detection, such as faster R-CNN, SSD, and R-FCN. However, directly applying these deep general object detection networks on license plate detection without modifying may not achieve good enough performance. This paper proposes a novel deep learning-based framework for license plate detection in traffic scene images based on predicted anchor region proposal and balanced feature pyramid. In the proposed framework, ResNet-34 architecture is first adopted for generating the base convolution feature maps. A balanced feature pyramid generation module is then used to generate balanced feature pyramid, of which each feature level obtains equal information from other feature levels. Furthermore, this paper designs a multiscale region proposal network with a novel predicted location anchor scheme to generate high-quality proposals. Finally, a detection network which includes a region of interest pooling layer and fully connected layers is adopted to further classify and regress the coordinates of detected license plates. Experimental results on public datasets show that the proposed approach achieves better detection performance compared with other state-of-the-art methods on license plate detection.

中文翻译：

---

用于交通场景图像中车牌检测的带有平衡特征金字塔的预测锚区域建议

车牌检测是智能交通系统中的关键问题。最近，已经提出了许多基于深度学习的网络，并在通用对象检测中取得了令人难以置信的成功，例如更快的R-CNN，SSD和R-FCN。但是，将这些深层通用对象检测网络直接应用于车牌检测而不进行修改可能无法获得足够好的性能。本文提出了一种基于深度学习的，基于预测锚区域建议和平衡特征金字塔的交通场景图像车牌检测框架。在提出的框架中，首先采用ResNet-34体系结构生成基本卷积特征图。然后使用平衡要素金字塔生成模块生成平衡要素金字塔，其中每个功能级别从其他功能级别获取相等的信息。此外，本文设计了具有新颖的预测位置锚定方案的多尺度区域投标网络，以生成高质量的投标。最后，采用包括感兴趣区域池化层和全连接层的检测网络，以进一步分类和回归检测到的车牌的坐标。在公共数据集上的实验结果表明，与其他最新的车牌检测方法相比，该方法具有更好的检测性能。采用包括感兴趣区域池化层和全连接层的检测网络对检测到的车牌的坐标进行进一步分类和回归。在公共数据集上的实验结果表明，与其他最新的车牌检测方法相比，该方法具有更好的检测性能。采用包括感兴趣区域池化层和全连接层的检测网络对检测到的车牌的坐标进行进一步分类和回归。在公共数据集上的实验结果表明，与其他最新的车牌检测方法相比，该方法具有更好的检测性能。