Output details

Articulated motion reconstruction from feature points

<23>Addresses problem of reconstructing non-rigid articulated motion from sequences of unstructured feature points (e.g., analysis of videos of human movement). The novelty of our model-based dynamic algorithm lies in its computational effectiveness and robustness, non-rigid pattern matching, and feature identification in the presence of missing and noisy data. The problem addressed is fundamental and crucial in vision-based motion tracking of complex objects. Our general approach of self-resuming tracking of articulated movements particularly improves the auto-labelling function in marker based motion capture systems, and has been successfully used in industrial motion capture animation/games projects, e.g. Dance:UK on Xbox.