Publication: Reconocimiento de acciones corporales mediante aprendizaje profundo con datos sensoriales de dispositivos móviles
Loading...
Official URL
Full text at PDC
Publication Date
2021-06-05
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
En la actualidad, el uso de redes neuronales recursivas es muy frecuente, aunque no lo veamos a simple vista. Desde nuestras pulseras inteligentes hasta la aplicación más sencilla, es posible que exista una de estas redes trabajando en un segundo plano. En este proyecto mostramos cómo, con una sencilla red, se puede llegar a clasificar movimientos corporales. Para ello se propone el uso de una red neuronal del tipo LSTM (Long Short Term Memory), de forma que utilizando datos de aceleración procedentes de los sensores integrados en un dispositivo móvil, es posible, por un lado entrenar el modelo de red indicado para proceder, en segundo lugar a la clasificación de nuevas acciones.
Se ha diseñado una aplicación que realiza esta funcionalidad. Además, se propone una extensión en el ámbito de lo que se conoce como Internet de las Cosas (IoT, Interner of Things) de forma que los resultados obtenidos se cargan en un servidor remoto, esto es, en la nube, que permite el acceso e intercambio de la información almacenada, mediante la red social Twitter. En el planteamiento que se propone se amplía la posibilidad de clasificar movimientos mediante una red previamente entrenada o bien diseñar una red definida ad hoc según las acciones requeridas.
Currently, the use of recursive neural networks is very frequent, although we don’t see it right away. From our smart bracelets to the simplest application, it is possible to find one of these networks working in the background. In this project, we show how with a simple network, it is possible to classify body movements. For this, the use of a neural network of the LSTM type (Long Short Term Memory) is proposed. It is feasible to use acceleration data from the sensors integrated in a mobile device and also to train the indicated network model to proceed to the classification of new shares. An application has been designed that performs this functionality. In addition, an extension is proposed in the field of what is known as the Internet of Things (IoT) so that the results obtained are uploaded to a remote server, that is, in the cloud, which allows access and exchange of stored information, through the social network Twitter. The suggested come up with the possibility of classifying movements through a previously trained network or designing a defined ad hoc network according to the required actions.
Currently, the use of recursive neural networks is very frequent, although we don’t see it right away. From our smart bracelets to the simplest application, it is possible to find one of these networks working in the background. In this project, we show how with a simple network, it is possible to classify body movements. For this, the use of a neural network of the LSTM type (Long Short Term Memory) is proposed. It is feasible to use acceleration data from the sensors integrated in a mobile device and also to train the indicated network model to proceed to the classification of new shares. An application has been designed that performs this functionality. In addition, an extension is proposed in the field of what is known as the Internet of Things (IoT) so that the results obtained are uploaded to a remote server, that is, in the cloud, which allows access and exchange of stored information, through the social network Twitter. The suggested come up with the possibility of classifying movements through a previously trained network or designing a defined ad hoc network according to the required actions.
Description
Trabajo de Fin de Grado en Ingeniería del Software , Facultad de Informática UCM, Departamento de Ingeniería del Software e Inteligencia artificial, Curso 2020-2021.