An Efficient and Layout-Independent Automatic License Plate Recognition System Based on the YOLO detector

This paper presents an efficient and layout-independent Automatic LicensePlate Recognition (ALPR) system based on the state-of-the-art YOLO objectdetector that contains a unified approach for license plate (LP) detection andlayout classification to improve the recognition results using post-processingrules. The system is conceived by evaluating and optimizing different models,aiming at achieving the best speed/accuracy trade-off at each stage. Thenetworks are trained using images from several datasets, with the addition ofvarious data augmentation techniques, so that they are robust under differentconditions. The proposed system achieved an average end-to-end recognition rateof 96.9% across eight public datasets (from five different regions) used in theexperiments, outperforming both previous works and commercial systems in theChineseLP, OpenALPR-EU, SSIG-SegPlate and UFPR-ALPR datasets. In the otherdatasets, the proposed approach achieved competitive results to those attainedby the baselines. Our system also achieved impressive frames per second (FPS)rates on a high-end GPU, being able to perform in real time even when there arefour vehicles in the scene. An additional contribution is that we manuallylabeled 38,351 bounding boxes on 6,239 images from public datasets and made theannotations publicly available to the research community.