Spatial Transform Depthwise Over-Parameterized Convolution Recurrent Neural Network for License Plate Recognition in Complex Environment

Abstract

Automatic license plate recognition system (ALPR) has been widely used in intelligent transportation and other fields. However, in complex environments such as sound source vehicle location, low light, or bad weather conditions, ALPR is still a challenging problem. Aiming at the problem, a deep learning framework is developed based on depthwise over-parameterized convolution recurrent neural network for license plate character recognition. The proposed framework is composed as follows: i) License plate correcting module based on Spatial Transformation Network; ii) Feature extraction module based on Depthwise Over-parameterized Convolution; iii) Sequence annotation module based on Bidirectional Long Short-Term Memory; iv) Regularized sequence decoding module based on Connectionist Temporal Classification with maximum conditional entropy. Two open-source datasets in China are used to verify the performance of the algorithm. The proposed framework can effectively correct distorted and inclined license plates in space, and recognize the license plate end-to-end, which avoids the complex character segmentation process. Compared with some current state-of-art algorithms, the proposed algorithm has higher recognition accuracy and better robustness.