Deep Learning e redes neurais convolucionais: reconhecimento automático de caracteres em placas de licenciamento automotivo

Abstract

Controlling road traffic, public or private, borders, parking lots, trespass traffic laws are examples of tasks that require license plate identification, being the license plates regulated by specific laws. For the license plate identification process, it is necessary the use of some computer vision technic, field that has gained enough attention of the scientific community and of the companies over the last years as a way to increase efficiency and to cut costs. With the advent of Deep Learning, solutions to the most diverse problems have been researched. With Artificial Neural Networks, especially of Convolutional Neural Networks, a jump in terms of results compared to traditional methods was given. Convolutional Neural Networks are extremely successful in tasks that mainly involve classification. This work aims to demonstrate the viability of a Convolutional Neural Network for reading characters in vehicular license plates, obtaining results on the order of 89.24% of correctly inferred characters.