License Plate Recognition System Based on Deep Learning

Abstract

We proposed a license plate recognition system based on the YOLOv2 framework. It contains a plate detection network and a plate recognition network. When we input a car image, the first network can detect the position of license plates, and the second network can recognize the characters within the detected plate. Besides, we modify the loss function to train the YOLOv2 for character recognition. Finally, the system outputs the predictive plate numbers after post-processing. Currently, the recognition accuracy can reach 97%, and at least 5 pictures can be recognized in one second.