Desarrollo de una aplicación móvil para detección de plazas disponibles y ocupadas en un estacionamiento utilizando aprendizaje profundo

Abstract

A parking lot is an area used by vehicle owners. When they go to look for a parking space and find it occupied, it becomes a problem, which is why the increase in vehicle congestion is a phenomenon that is seen more frequently in several areas, since most people have enough money to buy a vehicle. Some of the causes of traffic congestion are the large number of vehicles that circulate at certain times in the cities, and also the lack of parking spaces available in the parking lots, resulting in delays in the mobilization of people, increased pollution, traffic accidents, additional and excessive fuel consumption and even sometimes fines for non-compliance with the rules for parking in spaces not allowed. Smart parking uses technology that promises to provide drivers with better convenience when looking for a free parking space. For this reason, the present work focused on developing a mobile application capable of detecting parking spaces, integrating Deep Learning. For this, some images of the parking lot of the Faculty of Civil Engineering of the Technical University of Machala were taken for the development of the dataset, where some requirements were established, such as: the image must be of good quality, must have good resolution, must be images under different light conditions, among others. The images of the dataset were classified by classes, the first class corresponds to the images of free parking and the second class to occupied parking. The trained model was evaluated with a confusion matrix, which has the metrics of Precision, Recall and F-Score, in order to select the best trained model. A back-end containing the real-time detection part was also developed using the Visual Studio Code IDE and the Python programming language. On the other hand, for the development of the mobile application, the Android Studio IDE and the Java programming language were used, following the steps of the Mobile-D methodology, in which its phases were executed properly, such as the chronogram that allowed to have a better control of the activities during the development of the application. The design of the interface of the mobile application is simple for the user and intuitive with feedback, so that the user is able to quickly deduce how each of its buttons and sessions work. As a final result, we obtained a mobile application that allows to visualize, detect and identify the available and occupied parking spaces in the parking lot of the Faculty of Civil Engineering of the Technical University of Machala, in real time, applying Deep Learning. The tests were performed locally to verify the functions established in the requirements.