Viewer Preference of Aesthetic Pose Quality in a Gesture Corpus

This paper describes a computational aesthetics approach to inferring the quality of poses in a gesture corpus. We describe experimental design, feature implementation, and a pilot study. Motivation is taken from computational cinematography studies for virtual camera control. We apply similar methods to virtual character control, with the goal of enhancing avatar and NPC acting in cinematic videogames. Design includes Poserama, a gestural game designed to facilitate data collection to study character pose preferences by media consumers. Aesthetic features are based on arts and animation theory, and our algorithms score pose features along three dimensions: balance, asymmetry, and readability (BAR). The pilot study was run on a pre-processed corpus from motion capture data, with some encouraging results. This work is a first step in a computational perfomatology approach to skilled acting affordances in videogames, by learning the aesthetic features of affective pose representation in the arts.