Automatic Brazilian Mercosur license plate recognition: an approach with synthetic images

Abstract

Using images to obtain vehicle license information is not an exclusive of the present day. In 1987, Mizumachi already presented a system that performed the identification of numbers in vehicle plates. Systems such as these are presented to date, mainly through the feasibility of applications that can be carried out with these extractions, such as: identifying stolen or irregular vehicles, monitoring roads to precisely fine drivers, and controlling access to parking lots. The methods for solving this problem, generally referred as Automatic Plate Recognition (LPR) or Automatic License Plate Recognition (ALPR), consists in three steps: license plate identification in scenes, character segmentation and optical recognition of segmented characters. Although there are many solutions in the literature, the fact that there is no global standardization contributes to many proposals. Due to this lack of standardization, in 2014, Brazilian Resolution No. 510 established the system of vehicle identification plates in the Mercosur standard (CONSELHO NACIONAL DE TRÂNSITO (CONTRAN), 2014). With this new standard, existing solutions to the old standard do not behave as expected, mainly because of color changes, character shapes, and order of digits. So, in this project, a new method is proposed for the automatic recognition of vehicular plates of the Mercosur standard and a license plate embedding methodology for generating large volumes of Mercosur license plate images.