Emotional Movements Analysis and Classification with SRDA

Author

Wei Rang, Southern Illinois University CarbondaleFollow

Date of Award

8-1-2017

Degree Name

Master of Science

Department

Computer Science

First Advisor

Cheng, Qiang

Abstract

Motion databases are becoming larger with the development of hardware and software, applications based on these data have been widely used in many different areas, such as movies, animations, videos games, sports as well as computer vision and robotics. All those applications had made motion analysis and classification essential for better motion composition. However, in order to achieve a good connectivity between every motion and emotion behind it, it is totally important to understand human behavior, even if human movements are complex and hard to describe completely. In this thesis, we make investigations on connections between various emotional states and different movements with regards to the arousal and valence of the Russell's circumplex model. A variety of different features were applied to describe stylistic characteristics of motion based on Laban Movement Analysis(LMA). Motion capture data from dancers with various background were used for training and classification purpose. In our experiments, we have utilized four methods to finish multi-class classification: Random Forests(RF), Extremely Randomized Trees(ET), Support Vector Machines(SVM) and Spectral Regression Discriminant Analysis(SRDA). The experimental results show that extracted features based on LMA can provide a good description on emotion labels behind different motions. Furthermore, SRDA performed better than the other three classification methods.