Automatic Recognition for Arbitrarily Tilted License Plate

In this paper, we propose a novel automatic license plate recognition (ALPR) method based on convolutional neural network to achieve a better performance in detecting and recognizing license plate (LP) with relatively large angle of inclination. Most existing methods only perform well on dataset where LPs are presented in almost upright position with little or no tilted angle. While, in practice, the LP images collected by roadside cameras or hand-held image capturing devices can be fairly slanted, which causes great difficulties on recognition tasks. To solve this problem, we design an angle correction module and integrate it into a holistic ALPR model with a spatial transformer network embedded inside. The whole model can be trained end-to-end by back-propagation. A large and comprehensive rotated LP dataset Rlpd is collected and introduced in our work for model training and testing. Through extensive experiments, this approach is proved to have a better performance on tilted license plate dataset in terms of accuracy and computational cost than other state-of-the-art methods.