ETD

Archivio digitale delle tesi discusse presso l'Università di Pisa

Tesi etd-03012014-170133


Tipo di tesi
Tesi di dottorato di ricerca
Autore
ZARAKI, ABOLFAZL
URN
etd-03012014-170133
Titolo
Design and Experimental Evaluation of a Context-aware Social Gaze Control System for a Humanlike Robot
Settore scientifico disciplinare
ING-INF/06
Corso di studi
INGEGNERIA
Relatori
tutor Mazzei, Daniele
tutor Prof. De Rossi, Danilo
Parole chiave
  • Gaze behavior
  • Eye-tracking study
  • Attention system
  • Active vision
  • Social human-robot interaction
Data inizio appello
17/03/2014
Consultabilità
Completa
Riassunto
Nowadays, social robots are increasingly being developed for a variety of human-centered scenarios in which they interact with people. For this reason, they should possess the ability to perceive and interpret human non-verbal/verbal communicative cues, in a humanlike way. In addition, they should be able to autonomously identify the most important interactional target at the proper time by exploring the perceptual information, and exhibit a believable behavior accordingly. Employing a social robot with such capabilities has several positive outcomes for human society.
This thesis presents a multilayer context-aware gaze control system that has been implemented as a part of a humanlike social robot. Using this system the robot is able to mimic the human perception, attention, and gaze behavior in a dynamic multiparty social interaction.
The system enables the robot to direct appropriately its gaze at the right time to the environmental targets and humans who are interacting with each other and with the robot. For this reason, the attention mechanism of the gaze control system is based on features that have been proven to guide human attention: the verbal and non-verbal cues, proxemics, the effective field of view, the habituation effect, and the low-level visual features. The gaze control system uses skeleton tracking and speech recognition,facial expression recognition, and salience detection to implement the same features.
As part of a pilot evaluation, the gaze behavior of 11 participants was collected with a professional eye-tracking device, while they were watching a video of two-person interactions. Analyzing the average gaze behavior of participants, the importance of human-relevant features in
human attention triggering were determined. Based on this finding, the parameters of the gaze control system were tuned in order to imitate the human behavior in selecting features of environment.
The comparison between the human gaze behavior and the gaze behavior of the developed system running on the same videos shows that the proposed approach is promising as it replicated human gaze behavior 89% of the time.
File