Publication: Smart cities: Aprendizaje profundo con imágenes
Loading...
Official URL
Full text at PDC
Publication Date
2020
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
Con el paso del tiempo los sistemas de la información y la comunicación se han integrado en las sociedades humanas de tal forma que pasan desapercibidos en el entorno, para proporcionar, cada vez más y de forma más rápida, una gran cantidad de información que, al ser tratada, puede hacer que la vida de las personas sea más cómoda y sencilla.
Gracias a esta computación ubicua que monitoriza constantemente los datos de las actividades humanas han surgido paradigmas como Aprendizaje Profundo (Deep Learning) en Inteligencia Artificial o IoT (Internet of Things), que presta una especial atención a problemas del mundo real resolubles y automatizables, mediante el empleo de sistemas hardware y software que aplican los mejores modelos de procesamiento y análisis de datos para lograr una mejor toma de decisiones.
En particular, en el presente trabajo se pone el foco en el problema del análisis del flujo de vehículos existente en las ciudades modernas, que evolucionan hacia el concepto de ciudades inteligentes. Se trata de un problema relevante de cara a la sostenibilidad, un problema de la historia reciente que trata algunas cuestiones, como por ejemplo, la optimización de las rutas de transporte, la movilidad eficiente, la determinación de los flujos de transporte en función de los itinerarios de las ciudades, la gestión eficiente de las plazas de aparcamiento, entre otros aspectos relacionados.
Para abordar esta problemática se ha desarrollado un sistema software pensado para su implantación y uso en el mencionado contexto de las ciudades inteligentes.
Dicho sistema contiene las siguientes funcionalidades:
● Técnicas de análisis del movimiento en secuencias de imágenes mediante el cálculo del flujo óptico.
● Determinación de las regiones en movimiento mediante la aplicación técnicas de segmentación basadas en umbralización y etiquetado.
● Identificación de los vehículos en movimiento mediante técnicas basadas en Aprendizaje Profundo con Redes Neuronales Convolucionales.
● Desarrollo de un módulo de predicción del flujo de tráfico de vehículos con fines de prevención y planificación.
● Desarrollo de una arquitectura cliente-servidor para la implementación de métodos basados en el paradigma IoT, que permitan publicar y visualizar los datos en crudo y los resultantes del análisis del flujo de vehículos.
Over time, information and communication systems have been introduced into human societies in such a way that they are unnoticed in the environment to provide even more and more quickly, a lot of information that, when it’s processed, can make people's lives more simple and comfortable. Thanks to this ubiquitous computing that is permanently monitoring the data of human activities, paradigms such as Deep Learning in artificial intelligence or IoT have emerged, they pay special attention to real world problems that can be solved and automated by using hardware and software systems by applying the best data processing and analysis models to achieve a better decision making. Particularly, in the present work, the focus is put on the problem of vehicle flow análisis in modern cities, which are evolving to the concept of smart cities. This is a relevant problem in terms of sustainability, a problem of recent history that think about some questions, such as the optimization of transport routes, efficient mobility, determination of transport flows according to the timetables of the cities, efficient management of parking spaces, etc. To solve this problem we have developed a software system thought for it’s implantation and use in the context of smart cities. This system contains the following functionalities: ● Determination of regions in movement applying segmentation techniques based on umbralization and labeling. ● Movement analysis techniques applied to image sequences through the computation of optical flow. ● Vehicles identification in movement through techniques based on Deep Learning with Convolutional Neural Networks. ● Development of a prediction module for vehicle traffic flow with the aim of prevention and planification. ● Development of a client-server architecture for the implementation of methods based on the IoT paradigm, which allows to publish and to visualize raw data and the resulting data of vehicle flow analysis.
Over time, information and communication systems have been introduced into human societies in such a way that they are unnoticed in the environment to provide even more and more quickly, a lot of information that, when it’s processed, can make people's lives more simple and comfortable. Thanks to this ubiquitous computing that is permanently monitoring the data of human activities, paradigms such as Deep Learning in artificial intelligence or IoT have emerged, they pay special attention to real world problems that can be solved and automated by using hardware and software systems by applying the best data processing and analysis models to achieve a better decision making. Particularly, in the present work, the focus is put on the problem of vehicle flow análisis in modern cities, which are evolving to the concept of smart cities. This is a relevant problem in terms of sustainability, a problem of recent history that think about some questions, such as the optimization of transport routes, efficient mobility, determination of transport flows according to the timetables of the cities, efficient management of parking spaces, etc. To solve this problem we have developed a software system thought for it’s implantation and use in the context of smart cities. This system contains the following functionalities: ● Determination of regions in movement applying segmentation techniques based on umbralization and labeling. ● Movement analysis techniques applied to image sequences through the computation of optical flow. ● Vehicles identification in movement through techniques based on Deep Learning with Convolutional Neural Networks. ● Development of a prediction module for vehicle traffic flow with the aim of prevention and planification. ● Development of a client-server architecture for the implementation of methods based on the IoT paradigm, which allows to publish and to visualize raw data and the resulting data of vehicle flow analysis.
Description
Trabajo de fin de Grado en Grado en Ingeniería del Software , Facultad de Informática UCM, Departamento de Ingeniería del Software e Inteligencia Artificial, Curso 2019/2020