Facial Tracking for an Emotion-Diagnosis Robot to Support E-Learning

There have been a lot of researches on the detection/estimation of human emotions from facial expressions. However, most of them have extracted facial features for some specific emotions from the still pictures of artificial actions or performances. This paper describes facial tracking for e-learning support robot which can estimate a emotion of e-learning user from his/her facial expression in real-time; (1) the criteria of the facial expression to classify the eight emotions was obtained by the time sequential subjective evaluation on the emotions as well as the time sequential analysis of a facial expression by image processing. (2) The coincidence ratio between the discriminated emotions based upon the criteria of emotion diagnosis and the time sequential subjective evaluation on emotions for ten e-learning subjects was 69%. (3) Then, the possibility of the real time emotion diagnosis robot to support e-learning was confirmed by the facial image processing at the 15 frame/sec. rate as well as the simple emotion diagnosis algorithm based upon the Mahalonobis distance