Real-time tracking of multiple people using unlabelled markers and its application in interactive dance

This paper describes an approach to tracking the global locations of multiple dancers using marker-based motion capture system from only unlabeled 3D marker data for the application of interactive dance. This algorithm is based on the fact that in the presence of multiple people, marker labeling is poor or even infeasible, while reconstruction of the marker 3D coordinates is still reasonable. The tracking of dancers is done through treating dancers with markers as point clouds that have noticeable characteristics. Such characteristics include fixed marker set, all the markers being within an average wingspan, and that the markers on a dancer are closely bounded together. Using these characteristics, a blurring process was used to find the location of the dancers, and a mean shift algorithm was implemented to track them. Satisfactory results have been obtained using the proposed method.