We have developed novel techniques to build models of skeletal joints that are suitable for simulation. We can reconstruct polyhedral models from slices extracted directly from the CT data, or we can reconstruct (selectively) smooth models, represented as collections of polynomial surface patches, from unorganized large sets of points sampled on the model's surface.
In this application, we obtain the points by first extracting an
isosurface corresponding to the interface between bone and
surrounding
soft tissue.
We then segment the isosurface into its connected components,
corresponding to the various skeletal parts. The segmentation
process
currently requires human intervention to solve ambiguous cases.
We are
developing new tools to deal with this problem in a more
automated
way.
Finally, we use 
-solids [2] to ``clean up''
the data set from unwanted artifacts. The result is a
triangulated
two-manifols that can be used as a starting point for a smooth
model
reconstruction, as explained below.
In the rest of this section, we provide an overview of the algorithms used for the reconstruction tak.