A novel approach to posture recognition of ballet dance

The objective of this work is to recognize 17 fundamental ballet postures using a single camera. The proposed 7-stage algorithm first performs skin color segmentation on raw images and then finds minimized skeletons which are further approximated by straight lines using chain code and sampling. From the stick figures, the angles the significant lines make with the abscissa in dominant quadrants are determined. Finally, the unknown posture is recognized as the one of the 17 forms with which it holds maximum similarity as calculated by a similarity operator with average 88.9% accuracy.