Inferring performer skill from aesthetic quality features in a dance game gesture corpus
In this paper, we describe experiments for inferring the artistic skill of performers by analyzing pose features in a gesture corpus. Poses were generated by having participants play a popular Kinect dance game in a markerless motion capture studio. Skeletal data was analyzed for features derived from both statistical analysis as well as arts and animation theory, and aesthetic metrics were designed to score pose features along three dimensions: balance, asymmetry, and readability. We applied our metrics to poses in a corpus of 10,080 annotated frames generated from 20 dance performances ranked according to the performing arts background of each participant. This work is the foundation of a computational performatology approach to quantifying artistic gesture in media by identifying aesthetic features that indicate figurative quality to viewers. The potential application of gesture analysis and feedback will be to inform the design of performative logics for virtual control of avatars and non-player characters in videogames. Copyright © 2013, Association for the Advancement of Artificial Intelligence. All rights reserved.
