Desenvolvimento de um sistema móvel de reconhecimento de placas de veículos baseado em redes neurais artificiais

Abstract

In the last years, solutions based on computer vision have emerged thanks to the evolution of the processing capacity of computers and the techniques used, which have helped to implement these solutions even in tiny devices such as smartphones. Among the possible applications inside this field, there is the automatic recognition of vehicle license plates, used in parking lots and smart tolls, as well as in other situations, where a device receives or captures images of cars and recognizes the license plate of each one, obtaining its constituent characters. In this sense, this work aimed to develop an automatic license plate recognition system for mobile devices capable of integrating different applications. For this, computer vision algorithms and technologies, such as YOLO and Convolutional Neural Networks, were used together with public vehicle image datasets, such as the UFPR-ALPR dataset, to train models capable of detecting regions containing license plates, segment the characters inside the identified plates and recognize each one of those characters. In order to demonstrate a use case, the application used the identified vehicle plates to consult the status of each one, identifying if the vehicles referring to them are in a situation of theft or robbery. In the end, after evaluating the system’s performance, it was obtained 70,70% of success in identifying correctly the license plates from tests images set. That is, for every ten images containing a vehicle, seven license plates was correctly identified. To do this, the application spent, on average, three seconds to process each image.