The EMOSIGN - analyzing the emotion signature in human motion

We examine the relationship between human motion and emotions. In this paper, our purpose is to develop an analysis method which can be used to classify human motions associated with specific emotions. Our approach to motion data analysis is to apply higher order singular value decomposition (HOSVD) directly to motion data. HOSVD can generate a compact vector which specifies each emotion common among people. Experimentally, we obtained motion capture data for "gait" action related to six emotions - anger, joy, and so on. Human gait motion was also created with an animator. For these motion data, analysis showed that our method can classify the human motions specific to each emotion.