Publication: Reconocimiento de acciones de movimiento humanas con un smartphone mediante aprendizaje profundo
Loading...
Official URL
Full text at PDC
Publication Date
2022
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
En este trabajo se propone una aplicación inteligente con capacidad de procesar datos sensoriales procedentes de un dispositivo móvil, del tipo smartphone o tablet mediante técnicas basadas en Aprendizaje Profundo (Deep Learning). La finalidad consiste en la detección de acciones de movimiento humanas para su monitorización. Para ello se propone el uso de una red neuronal recurrente, cuyo modelo exacto es del tipo LSTM (Long Short Term Memory). Esta red está entrenada con datos de Aceleración, en los tres ejes (X, Y, Z), procedentes de los sensores del smartphone para reconocer cinco tipos concretos de acciones: “Dancing”, “Running”, “Sitting”, “Standing” y “Walking”. La aplicación está diseñada de tal manera que cuando se reciben datos de aceleración, correspondientes a una de las acciones mencionadas, éstos se procesan en la red para identificar el tipo de acción asociada. Además, de lo anterior, el trabajo posee una funcionalidad añadida bajo el paradigma de Internet de las Cosas (IoT, Internet of Things), de forma que los datos son enviados a una plataforma específica para IoT su almacenamiento junto con el resultado de su procesamiento mediante la red LSTM. Desde esta plataforma se pueden enviar mensajes sobre las acciones realizadas para ser recibidos en una cuenta de la red social Twitter, a la vez que es posible su visualización accediendo a la mencionada plataforma con fines de monitorización de las acciones almacenadas en ella. Por tanto, la aplicación desarrollada combina dos capacidades de interés, la primera se refiere a la utilización de técnicas inteligentes avanzadas mediante Deep Learning y la segunda a su extensión bajo el ámbito de IoT. Los experimentos realizados han verificado la validez de la propuesta, destacando los aspectos relacionados con el sistema inteligente y la navegación autónoma del vehículo autómata.
In this work, an intelligent application is proposed with the ability to process sensory data coming from a mobile device, such as a smartphone or tablet, and using techniques based on Deep Learning. The purpose is the detection of human movement actions for monitoring. A recurrent neural network is proposed for this purpose, whose exact model is of the type of LSTM (Long Short-Term Memory). This network is trained with acceleration data, in the three axes (X, Y, Z), from the smartphone sensors to recognize five specific types of actions: “Dancing”, “Running”, “Sitting”, “Standing” and “Walking”. The application is designed in such a way that when acceleration data are received, corresponding to one of the mentioned actions, they are processed in the network to identify the type of associated action. In addition to the above, a functionality, under the paradigm of Internet of Things (IoT, Internet of Things), is added, so that the data are sent to a specific IoT platform for storage together with the result of their processing through the LSTM net. This platform offers a offers a messaging service sending the type of action recognized and stored, which can be received through a Twitter social network account, while it is possible to gain access to the platform for the purposes of monitoring the actions stored in it. Therefore, the developed application combines two abilities of interest, the first refers to the use of advanced intelligent techniques through Deep Learning and the second to its extension under the scope of IoT.
In this work, an intelligent application is proposed with the ability to process sensory data coming from a mobile device, such as a smartphone or tablet, and using techniques based on Deep Learning. The purpose is the detection of human movement actions for monitoring. A recurrent neural network is proposed for this purpose, whose exact model is of the type of LSTM (Long Short-Term Memory). This network is trained with acceleration data, in the three axes (X, Y, Z), from the smartphone sensors to recognize five specific types of actions: “Dancing”, “Running”, “Sitting”, “Standing” and “Walking”. The application is designed in such a way that when acceleration data are received, corresponding to one of the mentioned actions, they are processed in the network to identify the type of associated action. In addition to the above, a functionality, under the paradigm of Internet of Things (IoT, Internet of Things), is added, so that the data are sent to a specific IoT platform for storage together with the result of their processing through the LSTM net. This platform offers a offers a messaging service sending the type of action recognized and stored, which can be received through a Twitter social network account, while it is possible to gain access to the platform for the purposes of monitoring the actions stored in it. Therefore, the developed application combines two abilities of interest, the first refers to the use of advanced intelligent techniques through Deep Learning and the second to its extension under the scope of IoT.
Description
Trabajo de Fin de Grado en Ingeniería del Software, Facultad de Informática UCM, Departamento de Ingeniería del Software e Inteligencia Artificial, Curso 2021/2022.