车牌检测识别--Towards End-to-End Car License Plates Detection and Recognition with Deep Neural Networks

Towards End-to-End Car License Plates Detection and Recognition with Deep Neural Networks
https://arxiv.org/abs/1709.08828

本文使用CNN网络来进行车牌检测和识别，一个网络完成端对端训练，检测和识别不用分离。
The whole framework takes 0.3 − 0.4 second per image on a Titan X GPU

网络流程如下所示：

首先使用 VGG-16 中的卷积层来提取特征，然后使用 Region Proposal Network (RPN) 来提取车牌候选区域，有了车牌区域，我们进行车牌识别，这里我们没有进行车牌分割， regard the plate recognition as a sequence labeling problem。

Bidirectional RNNs (BRNNs) with CTC loss [14] are employed to label the sequential features

识别效果图：

Performance Evaluation on PKUData

11