Emotion recognition from 3D videos using optical flow method

Facial expressions provide crucial clues about emotions. Emotions are signalized as reactions to the events of a person. The way people think, behave and act are controlled by emotions. Recognition of emotions via facial expressions from 3D videos plays a crucial role in various disciplines where essential key changes in facial movements are to be identified rapidly. In this paper, an approach using combination of geometric feature based approach and optical flow method is proposed for recognizing six basic emotions from video sequences of BU-4DFE database and in-house developed video streams. Out of 83 facial points given in the BU4DFE database, 25 optimum feature points are considered. The optical flow is estimated between the facial points in neutral and apex frames and by convolution of matrices feature vectors are constituted. The feature vectors that are obtained for all the subjects of various emotions are fed to Neural Networks (NN) for classification. The method has been analyzed on BU4DFE database and in-house developed videos. The proposed method resulted in accuracy comparable with literature.