Publication: Integración de servicios y tecnologías IoT: detección de vehículos y predicción de niveles de tráfico
Loading...
Official URL
Full text at PDC
Publication Date
2019
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
Las congestiones de tráfico vehicular son un problema que sufren muchas ciudades en el mundo, cada día es mayor la cantidad de personas que pierden valiosas horas de su tiempo atascadas en el tráfico, lo que se traduce en pérdidas económicas y de otra índole. Una posible solución para reducir el tiempo perdido de los viajeros por congestión y la contaminación creada por los estancamientos, es alertar acerca del flujo de tráfico presente en las vías, de modo que un conductor pueda escoger su ruta conociendo previamente el tráfico existente por zonas. El presente trabajo se focaliza en una solución para monitorizar el flujo vehicular en las vías con el fin de que estos datos sean accesibles por parte de los usuarios, de igual modo, podrán obtener información sobre el tráfico estimado en cierto momento de cualquier día para dichas vías. Esta solución está basada en la integración de servicios y tecnologías de Internet de las Cosas, IoT (Internet of Things), y además está pensada para ser totalmente escalable, de manera que sirva para sensorizar no solo una sino todas las vías que se deseen. El planteamiento de la solución consiste en el uso de cámaras inteligentes con procesamiento de borde (edge computing) que implementan contenedores Docker, éstos son capaces de ejecutar códigos en Python con modelos de redes neuronales entrenadas con TensorFlow para reconocimiento de vehículos, luego este código envía por MQTT a Azure IoT Hub los resultados del análisis y, posteriormente, estos datos se almacenan en una base de datos MongoDB hospedada en Azure. Una vez recolectados estos datos, una función Lambda programada en Python y publicada en AWS, a través de Amazon API Gateway, permite actuar como API Rest para consultas HTTP desde múltiples dispositivos y aplicaciones, que sirven para preguntar el nivel de tráfico actual o estimado en cierta fecha y hora. Para predecir el tráfico en otra fecha se utiliza un modelo de regresión lineal realizada con Apache Spark a partir de los datos obtenidos. En este trabajo se explica el proceso evolutivo de todas las pruebas y cambios realizados para llegar a diseñar la solución que se propone, junto con las evidencias de su viabilidad en base a los resultados obtenidos.
Traffic congestion is a problem that many cities in the world, every day, more people lose valuable hours of their time stuck in traffic, which translates into economic and other losses. A possible solution to reduce the lost time of the travelers due to congestion and the pollution created by the vehicular stagnations is to warn about the traffic flow present on the roads, so that a driver can choose his route knowing previously the existing traffic by zones. The present work focuses on a solution to monitor vehicular flow on the roads so that these data are accessible by users, in the same way, they can obtain information on the estimated traffic at a certain time of any day for such roads. This solution is based on the integration of services and technologies of Internet of Things, IoT, and is also designed to be fully scalable, so that it serves to sensorize not only one, but all the desired routes. The solution is based on the use of intelligent cameras with edge processing (edge computing) that implement Docker containers, they are able for executing Python codes with models of neural networks trained with TensorFlow for vehicle recognition, then this code sends by MQTT to Azure IoT Hub the results of the analysis and, subsequently, this data is stored in a MongoDB database hosted in Azure. Once this data has been collected, a Lambda function programmed in Python and published in AWS, through Amazon API Gateway, allows acting as a Rest API for HTTP queries from multiple devices and applications, which are used to ask the current or estimated traffic level at an specific date and time. To predict traffic on another date, a linear regression model using Apache Spark is used based on the obtained data. This work explains the evolutionary process of all the tests and changes done in order to get the design of the proposed solution, along with the evidence of its feasibility based on the results obtained.
Traffic congestion is a problem that many cities in the world, every day, more people lose valuable hours of their time stuck in traffic, which translates into economic and other losses. A possible solution to reduce the lost time of the travelers due to congestion and the pollution created by the vehicular stagnations is to warn about the traffic flow present on the roads, so that a driver can choose his route knowing previously the existing traffic by zones. The present work focuses on a solution to monitor vehicular flow on the roads so that these data are accessible by users, in the same way, they can obtain information on the estimated traffic at a certain time of any day for such roads. This solution is based on the integration of services and technologies of Internet of Things, IoT, and is also designed to be fully scalable, so that it serves to sensorize not only one, but all the desired routes. The solution is based on the use of intelligent cameras with edge processing (edge computing) that implement Docker containers, they are able for executing Python codes with models of neural networks trained with TensorFlow for vehicle recognition, then this code sends by MQTT to Azure IoT Hub the results of the analysis and, subsequently, this data is stored in a MongoDB database hosted in Azure. Once this data has been collected, a Lambda function programmed in Python and published in AWS, through Amazon API Gateway, allows acting as a Rest API for HTTP queries from multiple devices and applications, which are used to ask the current or estimated traffic level at an specific date and time. To predict traffic on another date, a linear regression model using Apache Spark is used based on the obtained data. This work explains the evolutionary process of all the tests and changes done in order to get the design of the proposed solution, along with the evidence of its feasibility based on the results obtained.
Description
Trabajo de Fin de Máster, Universidad Complutense, Facultad de Informática, Departamento de Ingeniería del Software e Inteligencia Artificial, Curso 2018/2019