Emotion Recognition from Human Gait Features Based on DCT Transform

Emotion recognition is of great value in human-computer interaction, psychology, etc. Gait is an important pattern of emotion recognition. In this paper, 59 volunteer’s gait data with angry or happy emotion, have been collected by the aid of Microsoft Kinect. The gait data are treated as discrete time signals, and we extract a series of frequency features based on the discrete cosine transform. Simultaneously, we have established emotion recognizing models with SVM, the K-nearest neighbors, and decision tree. The best recognition rate can exceed 80%, which indicates that our proposed features are useful for recognizing emotions.