Feature fusion based automatic aesthetics evaluation of robotic dance poses

Inspired by human dancers who make a comprehensive aesthetic judgement of their own dance poses by using both visual and non-visual information, this paper presents a novel feature fusion based approach to automatic aesthetics evaluation of robotic dance poses in order to improve the performance of robotic choreography creation. Four kinds of features are extracted, namely kinematic feature, region feature, contour feature, and spatial distribution feature of colour block. Based on different feature combinations, machine learning is deployed to train aesthetics models for the automatic judgement on robotic dance poses. The proposed approach has been implemented on a simulated robot environment, and experimental results are presented to verify its feasibility and good performance.