An Edge-end Based Fast Car License Plate Recogniton Method

For the size of model file in mainstream car license plate recognition methods are too large, they are unsuitable for deployment on edge end. To solve this problem, a fast license plate recognition method on embedded platforms for edge-end is proposed. In this method, SSD-MobileNet is adopted to detect the license plate firstly. Then, an end-to-end method without character segmentation is used to recognize license plate character, which can avoid the problem that the wrong license plate character segmentation affects license plate recognition. The trained model size in this method is 1.9MBit, and the experimentation show that the proposed method reaches a high accuracy and fast speed in both detection and recognition.