Skeleton Construction from Motion Capture Data

Gleicher, M. A Simple Motion Processor – DRAFT.
BibTeX entry
Another chapter from the Gleicher's intended book. Thorough description of a simple geometric approach to reconstruct skeletal angles from a captured set of markers. The processed assumes clean marker data and a skeleton to be given. An exact sequence of steps must be written by human to construct skeleton joints from a particular marker set. The angles are computed from joint positions then.
O'Brien, J. – Bodenheimer, R. – Brostow, G. – Hodgins, J. Automatic Joint Parameter Estimation from Magnetic Motion Capture Data. Graphics Interface. 2000.
Construction of articulated structure (skeleton) from magnetic motion capture data. Joint positions, limb lengths or even hierarchy are automatically computed using least squares fitting. For each pair of rigid limbs the joint is computed as a point with (almost) constant coordinates in both coordinate systems of the limbs. Magnetic MC data are required because the orientation of the limbs and hence the orientation of the markers is needed.
Orientation of the limbs could be obtained from several markers on each rigid limb, using any motion capture system (e.g. optical), not only the magnetic one. These markers determine a coordinate system and orientation of the limb. The groups of markers atached to particular limbs may be computed automatically or given as input. It still presents a fraction of necessary work compared to creation of a complete skeleton usually assumed to be given before conversion of MC data to skeletal data. The skeleton creation requires very tedious and inaccurate limb lenghts measurement.
Automatic detection of markers atached to one limb: if a pair of markers is in a fixed distance during the whole motion. (?)
And what about estimation of joint angle ranges?
Silaghi, M-C. – Plänkers, R. – Boulic, R. – Fua, P. – Thalmann, D. Local and Global Skeleton Fitting Techniques for Optical Motion Capture. Modelling and Motion Capture Techniques for Virtual Enviroments, volume 1537 of Lecture Notes in Artificial Intelligence. 1998.
Taycher, L. – Fisher, J. W. – Darrell, T. Recovering Articulated Model Topology from Observed Motion.
A brief and obscure draft about using probabilistic methods and "scaled prism model" to recover articulated topology of a figure in motion. Rigid segments are most likely to be connected if their motion is dependent, i.e. entropy of their relative motion is small. The minimal entropy spanning tree gives the articulated hierarchy then. I didn't understand it very well.
Update: 20. 10. 2004