Artificial Mercosur License Plates
Description
In 2018, a new Mercosur license plate standard was published, unifing the identification of vehicles and replacing the present license plate standards of five South America countries. In this new scenario, automatic license plate recognition (ALPR) systems built upon supervised learning algorithms could not be trained due to the lack of available data in real scenarios. So, in order to create a dataset without real samples of the new license plates standard, enabling the trainment of these models, a Mercosur license plate generator was developed to generate artificial license plates images with shadow, occlusion and other variations to mimetize real conditions, and a embeding system with license plate detection (LPD) that detects old 3-letter license plates in images of real scenarios and overwrite it with an artificially generated license plates. The dataset contains images of real scenarios where 3-letter license plates were detected using YOLOv3 (http://arxiv.org/abs/1804.02767) and overwritten by artificially generated images of the new mercosur license plates. It is organized in two folders: images - containing the images (JPEG) of the dataset; and labels - containing text files with the class identification number and the coordinates of the detected license plates in the image, following the Yolo_mark annotation specification (https://github.com/AlexeyAB/darknet#how-to-train-to-detect-your-custom-objects). The images are separated in five classes, identified by a prefix in the filename: 1) monitoring_system_ - 2925 images acquired by a license plate detection model applied in a public videomonitoring system; 2) parking_lot1_ - 566 images of cars in a parking lot acquired using a smartphone camera; 3) parking_lot2_ containing 23 images of the same as parking_lot1_ but acquired using a tablet camera; 4) parking_lot3_ 11 images same as parking_lot1_ but acquired using a different smartphone model; and 5) cropped_parking_lot containing 315 images cropped in the license plate area from parking_lot_ images; Also, there is a CSV file listing all license plates detected in all images, organized in seven columns: image, label, class, x_center, y_center, width and height. The image column is the filename of the image containing the license plate, label is the filename of its recpective annotation, class is the class of the object (in this case, always zero, the index of the license plate object), and x_center, y_center, width and height, the coordinates of the rectangle of the set of pixels representing the license plate in the image, following YOLO annotation standard.
Files
Steps to reproduce
The steps to reproduce this dataset can be found in the paper Brazilian Mercosur License Plate Detection: a Deep Learning Approach Relying on Synthetic Imagery (https://ieeexplore.ieee.org/document/9046091).