License Plate Recognition Algorithms for Complex Situations

Under ideal conditions, license plate recognition accuracy is high. However, the accuracy rate decreases in complex scenes, such as tilted license plate, weather effects, long distance and multiple license plates. In this paper, the EAST model for text detection in complex scenes is improved to address this problem, and the attention mechanism is introduced to improve the sensitivity to license plate information. Meanwhile, by improving the loss function, it learns to judge the relative position information of real and predicted frames to improve the detection accuracy. In addition, the spatial transformation network STN is introduced into the text recognition model CRNN to correct the distortion on the license plate and improve the generalization ability of the model. Experiments show that our improved license plate detection and recognition algorithm outperforms the current mainstream algorithms.