Facial emotion recognition using depth data

In this paper an original approach is presented for facial expression and emotion recognition based only on depth channel from Microsoft Kinect sensor. The emotional user model contains nine emotions including the neutral one. The proposed recognition algorithm uses local movements detection within the face area in order to recognize actual facial expression. This approach has been validated on Facial Expressions and Emotions Database using 169 recordings of 25 persons. Though an average recognition accuracy is slightly above 50% this approach is highly independent of illumination conditions and also accepts low distance between sensor and the user. Thus, the proposed approach can be used to support other algorithms based on optical channel, as well as using skeleton or face tracking information.