Abstract
In order to have a good communication between human and robot, it is important for robot to infer the kind and the degree of human emotions from his/her behaivior. This paper proposes a human-like recognition system to identify the degree of human emotions from beckon motion. Three kinds of emotions are selected to be recognized, which are “Angry”, “Hurry” and “Friendly”. First, using the Subjective Rating method, human ability is analyzed of estimating others' mental states from some kinds of emotional beckon motions. This result shows that human can distinguish beckon motions which reflect three different kinds of emotions. Second, six feature parameters are extracted from a hand-tip motion to describe the characteristics of each emotional beckon movements. Some of these parameters have correlationship to the degree of emotions evaluated by the Subjective Rating method. Then, human-like recognition system is constructed with neural network. This neural network learns the relationship between six feature parameters and the results of the Subjective Rating. Finally the validity of this recognition system is confirmed by experiments.
