Técnicas de aprendizagem profunda para estacionamentos inteligentes.

Abstract

Parking in big cities has become a huge challenge, as the growing number of private cars purchased in recent decades contributes to the lack of urban mobility. Drivers often waste far more time than necessary to park their cars, contributing to traffic stress. Artificial intelligence has been looking for methods to solve this problem, this has only become possible nowadays due to the availability of large data resources and an improvement in the processing capacity of computers. A system of vision technique with a strong trend in recent years is deep learning with convolutional network algorithms. This work aims to evaluate deep learning techniques for the construction and evaluation of a model for intelligent parking, using a dataset from UFPR, in three distinct scenarios. The model performs the prediction in real-time and proposes a better cost-benefit ratio than alternatives with the Internet of Things, eliminating changes in the environment and being easy to maintain.