Classification of Emotions Indicated by Walking Using Motion Capture

Abstract

Research on the classification of emotions includes research using brain waves and heartbeats. However, in each case, the requirement to wear devices to take measurements is burdensome for subjects. Therefore, the purpose of this study is to classify emotions by analyzing walking. We propose an emotional analysis method that can analyze emotions numerically. The proposed linear model is composed of three matrices: an emotional matrix, A; an emotional vector, Z; and a biological vector, C. The emotion vector represents a subjective value of emotion, and the biological vector represents measured biological data. The emotion matrix converts the biological vector into an emotional vector. Therefore, the linear model is represented by Z = AC. Ten sets of walking episodes were measured per person. The first five sets are data for learning, and the second five sets are data for classification. The subjects listened to classical music and quantified their emotions using questionnaires. Five gaits were used for classification: stride length, arm amplitude, speed, foot height, and hand height. The highest accuracy rate in the classification of emotions under the emotional analysis method was 80%. Analysis of data from the walking experiments revealed that subjects with a high classification accuracy rate showed emotions while walking. On the other hand, subjects with a low classification accuracy rate did not show emotions while walking. Since the maximum difference was as large as 60%, it is considered that the ease of expressing emotions greatly affects the classification accuracy rate. It was suggested that the classification of emotions using the emotional analysis method is effective for people who tend to express emotions in their walking style. Future tasks include proposing new analytical methods, examining more suitable gaits, and classifying emotions into more categories.