A carregar...
Publicação
EmoStatus - Estimating the emotional state of people using electrodermal activity
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 826.83 KB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Emotional recognition is an area with growing importance, with applications in areas such as medicine, advertising and even software design. Electrodermal Activity is one of the physiological signals most used to perform emotional recognition. There are many ways researchers use this signal to predict emotions, but generally they use a small set of emotions, are not concerned with the speed of the algorithm, and very few look into the differences between men and women. As such, this work intends to develop an algorithm that can predict any emotion in real-time and to determine if separating data from men and women improves the results. To do so, we studied the current methods for emotion recognition and chose the ones that best fit our purposes in terms of speed and accuracy. We also identified the common general steps that most researchers use for emotion recognition. With algorithmic speed in mind, and with the knowledge obtained from the research, we built a general purpose emotional recognition framework which uses small blocks that communicate amongst each other and execute in paralell, removing any possible delay in the estimation thus allowing real-time estimation. We implemented our algorithm using this framework. Experimental evaluation showed that our algorithm achieves estimations with very small errors in the AMIGOS dataset and an accuracy for the estimation of quadrants of 96% for both genders, 97% for males and 94% for females. For the DEAP dataset, values of 82% for both genders and 85% for males and females were achieved. When compared with existing works, our algorithm presents better results, both for the estimation of valence and arousal and for the estimation of the quadrants. Finally, our algorithm performs its estimations in under 10ms, therefore it can be used for real-time experiments.
Descrição
Tese de mestrado, Engenharia Informática (Engenharia de Software) Universidade de Lisboa, Faculdade de Ciências, 2019
Palavras-chave
Reconhecimento de emoções Tempo real Excitação Valência Atividade eletrodérmica Teses de mestrado - 2019