ETD

Archivio digitale delle tesi discusse presso l'Università di Pisa

Tesi etd-03272018-110952


Tipo di tesi
Tesi di laurea magistrale
Autore
SEMERARO, FRANCESCO
URN
etd-03272018-110952
Titolo
Emotion recognition through physiological wearable sensors for social robotics applications using machine learning techniques
Dipartimento
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
BIONICS ENGINEERING
Relatori
relatore Dott. Cavallo, Filippo
controrelatore Prof. Vanello, Nicola
correlatore Dott.ssa Fiorini, Laura
Parole chiave
  • affective computing
  • system integration
  • mood induction procedure
  • sensor fusion
  • supervised learning
Data inizio appello
03/05/2018
Consultabilità
Completa
Riassunto
Nowadays, social robotics is a very recent topic of research. Aiming for futuristic scenarios, its objective is to introduce robots in the everyday life and making them interact with people in a way that resembles that one a normal human would perform. In this context, ability for robots to perceive human emotions and be consequently affected by them surely constitutes a crucial element in the building of human-robot interactions.
This work aims to assess the possibility for robots to recognize user’s emotions in social environments by means of physiological signals. Specifically, brain, cardiac and electrodermal activities were investigated through three different sensors, supported by video acquisition. To control all of them at the same time, a customized interface was implemented in Microsoft Visual Studio™ 2017 (Microsoft®, Washington, USA).
Secondly, an experimental methodology to elicit three emotional states by means of social interactions was designed. Paradoxically, this is the least explored among mood induction procedures, although it is one of the most natural and immediate stimuli to elicit emotions. 21 subjects were examined. The experiment succeeded in arousing desired emotional responses.
After the experiment, acquired signals were processed, and features were extracted and selected by means of correlation and significance analysis in Matlab™ 2017(Mathworks®, Massachusetts, USA). Finally, the obtained dataset was analyzed by different classifiers in WEKA™ 3.8 (University of Waikato, Waikato, New Zealand) framework, achieving 89.70% accuracy in discerning among a positive, negative and relaxed mood, in the best case. Different combinations of used sensors were considered in the analysis.
File