Towards end-to-end car license plate location and recognition in unconstrained scenarios

Benefiting from the rapid development of convolutional neural networks, performance car license plate detection and recognition has been largely improved. Nonetheless, most existing methods solve problems separately, focus on specific scenarios, which hinders deployment for real-world applications. To overcome these challenges, we present an efficient accurate framework to tasks simultaneously. It is a lightweight unified deep network, that can be optimized end-to-end work in real-time. Specifically, unconstrained anchor-free method adopted efficiently detect bounding box four corners plate, are used extract rectify target region features. Then, novel network branch designed further features characters without segmentation. Finally, task treated as sequence labeling problems, solved by Connectionist Temporal Classification (CTC) directly. Several public datasets including images collected different scenarios under various conditions chosen evaluation. Experimental results indicate proposed significantly outperforms previous state-of-the-art both speed precision.