In this paper, while performing functional analysis of the visual reaction which appears during a dialog for the purpose of realizing a more advanced dialog system, the detection technique of the head gesture which bears an important role there is proposed. First, in order to analyze the visual reaction under dialog, face-to-face human dialogue was filmed. For each frame of picture in the filmed data, the head area was extracted using color information. The optical flow in the area and it utilized as the feature vector for the gesture recognition was calculated. HMM based gesture recognition is applied to spot three kinds of head gesture, "nodding", "doubting", and "shaking". The unification method of voice information was also investigated in order to improve recognition performance.
展开▼