Enhancing motion capture performance by means of an internal anthropometric skeleton model

Most motion tracking algorithms rely on an initial skeleton model that has already been fitted to a special posture setup. However, such a first identification of markers often requires multiple manual actions of a designer. To automate this process, a novel approach for adapting a basic skeleton model to empirical motion capture data is presented. The approach is based on the anthropometric dimensions of a subject and subsequent tree-based skeleton fitting. It generates a tree representation of different possible skeleton configurations. The tree is annotated with costs based on discrepancies between markers and anatomic landmarks. A computation of the least cost path through the tree automatically results in an optimal fitting of the observed markers to the given anthropometric data of the subject.