DISEÑO DE UN ALGORITMO DE RECONOCIMIENTO DE PLACAS VEHICULARES ECUATORIANAS USANDO REDES NEURONALES CONVOLUCIONALES

Abstract

Detecting stolen vehicles in Ecuador is one of the important tasks for the country’s security agencies. One of the ways to increase the detection and recovery rate of vehicles is through intelligent applications, based on artificial intelligence and neural networks. These applications allowed the development of new vehicle license plate recognition (LPR) techniques. In this article, we propose a model to detect moving car plates using convolutional neural networks. First, a tagged manual is made on the images of the vehicle, identifying the location of the plate and the characters within the image. This information is presented in a convolutional neural network architecture for training and testing. The network architecture developed by Google, COCO Inception V2, is used as a training base, where we get our own trained model for Ecuadorian vehicles plates. The experimental results find that the present work achieves a favorable recognition precision of 85.1 in terms of the Ecuadorian plate training photo data set. It is worth specifying the exclusive use of open source software for the development of this work.