Abstract

Automatic License Plate Recognition (ALPR) has been a frequent topic ofresearch due to many practical applications. However, many of the currentsolutions are still not robust in real-world situations, commonly depending onmany constraints. This paper presents a robust and efficient ALPR system basedon the state-of-the-art YOLO object detector. The Convolutional Neural Networks(CNNs) are trained and fine-tuned for each ALPR stage so that they are robustunder different conditions (e.g., variations in camera, lighting, andbackground). Specially for character segmentation and recognition, we design atwo-stage approach employing simple data augmentation tricks such as invertedLicense Plates (LPs) and flipped characters. The resulting ALPR approachachieved impressive results in two datasets. First, in the SSIG dataset,composed of 2,000 frames from 101 vehicle videos, our system achieved arecognition rate of 93.53% and 47 Frames Per Second (FPS), performing betterthan both Sighthound and OpenALPR commercial systems (89.80% and 93.03%,respectively) and considerably outperforming previous results (81.80%). Second,targeting a more realistic scenario, we introduce a larger public dataset,called UFPR-ALPR dataset, designed to ALPR. This dataset contains 150 videosand 4,500 frames captured when both camera and vehicles are moving and alsocontains different types of vehicles (cars, motorcycles, buses and trucks). Inour proposed dataset, the trial versions of commercial systems achievedrecognition rates below 70%. On the other hand, our system performed better,with recognition rate of 78.33% and 35 FPS.