Automatic quality assessment system for attack motions using support vector machine

Abstract

In this study, we developed an automated system to evaluate quality of attack motions, which mimics director’s subjective assessment. Our method learns a support vector machine from a collection of history data of artists’ work to distinguish the final version of the motion asset and the motion in the middle of the production. We introduce a numerical feature to quantity ‘reservoir’ and ‘divergence’ which are assumed to be an important characteristics of action motion in video games.