Publication: Sistema de detección de emociones para su aplicación en navegación semántica
Loading...
Identifiers
Publication date
2019-07-08
Defense date
2019-07-08
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Las emociones son una parte muy importante en el ser humano, afectan en gran medida a las personas y a menudo se toman decisiones guiados por estas. Es por ello que contienen una información muy valiosa acerca de las personas y su entorno. Debido al valor de esta información vamos a desarrollar un sistema automático para la detección de emociones y se analiza como expresamos las personas las emociones.
Para el desarrollo del sistema se va a hacer uso de las técnicas de visión artificial y aprendizaje automático para tratar de predecir las emociones mediante el análisis de la expresión facial. Además, ha sido implementado el sistema en ROS para su posible aplicación en robótica. Una vez desarrollado el sistema se compara el rendimiento de distintos algoritmos de aprendizaje tales como Máquinas de Soporte Vectorial, Bosques Aleatorios, K-Vecinos y Perceptrón Multicapa y se analiza la forma en la que el sistema obtiene la información sobre la emoción; para ello es medida la influencia de las distintas zonas faciales con métricas típicamente usadas en el aprendizaje automático.
Finalmente se obtiene un sistema con una precisión del 84% de acierto empleando una Máquina de Soporte Vectorial entrenada para detectar 5 emociones: alegría, tristeza, ira, sorpresa y neutra. Se comprueba que la Maquina de Soporte Vectorial es un algoritmo que ofrece una mayor precisión para este problema por su planteamiento y por el conjunto de datos disponible para el entrenamiento. Además, se observa que ciertas zonas faciales como la boca o los ojos proporcionan mayor información que otras como pudiera ser la nariz.
Emotions are one of the most important features of human psychology and behavior, they have information about ourselves as well as our surroundings that can be useful for certain tasks in robotics. That is the reason why we developed a system for the detection of emotions based on the facial expression analysis. For the development of the system we made use of machine learning techniques that made it possible to build the system just with some examples of the same type of data we want to predict. We also implemented the system in ROS for a possible application in robotics related with semantic navigation. For the development first the data was extracted and organized. Once the data is ready a feature selection is made as well as a parameter tuning for getting the most out of the learning process. After all this is done, learning takes place and the implementation in ROS is done once it has finished. To determine the importance of each facial zone, a study was carried out by looking at the performance of the system when we do not use the information provided by that facial zone and when just that information is available for the learning algorithm. A comparison of different learning algorithms was also done by comparing the accuracy of four different algorithms: SVM, KNN, Random Forest and MLP, this study is not only sensitive to the type of problem but also the approach it is taken as well as the data available to us. Finally, a system with a 84% precision was obtained with the use of an SVM which fulfils with the objective of this project, but also leaves lots of new possible improvements to be done in the future that can result in a better performance and new functionalities.
Emotions are one of the most important features of human psychology and behavior, they have information about ourselves as well as our surroundings that can be useful for certain tasks in robotics. That is the reason why we developed a system for the detection of emotions based on the facial expression analysis. For the development of the system we made use of machine learning techniques that made it possible to build the system just with some examples of the same type of data we want to predict. We also implemented the system in ROS for a possible application in robotics related with semantic navigation. For the development first the data was extracted and organized. Once the data is ready a feature selection is made as well as a parameter tuning for getting the most out of the learning process. After all this is done, learning takes place and the implementation in ROS is done once it has finished. To determine the importance of each facial zone, a study was carried out by looking at the performance of the system when we do not use the information provided by that facial zone and when just that information is available for the learning algorithm. A comparison of different learning algorithms was also done by comparing the accuracy of four different algorithms: SVM, KNN, Random Forest and MLP, this study is not only sensitive to the type of problem but also the approach it is taken as well as the data available to us. Finally, a system with a 84% precision was obtained with the use of an SVM which fulfils with the objective of this project, but also leaves lots of new possible improvements to be done in the future that can result in a better performance and new functionalities.
Description
Keywords
Reconocimiento facial, Detección facial, Emociones, Algoritmos, ROS (Robot Operating System), Aprendizaje automático