Motion Correction of Interactive CG Avatars Using Machine Learning

Motion capture allows users to control their CG avatar via their own movements. However, the composed avatar motion fails to deliver the actual input movements if the user's motion information is not accurately captured due to measurement errors. In this paper, we propose a method that complements a user's motion according to the motion of another person for a two-party motion with interaction. This method is expected to compose avatar motions that look natural to the other person while emphasizing the actual motions of the user.