Modeling and Animating Human Figures - Computer Animation: Algorithms and Techniques - page 284

smoothly curving surface. It might require hundreds or thousands of polygons to achieve the same
visual quality as could be obtained with a single NURBS patch.
Patch representations
Virtual humans constructed with an emphasis on visual quality are frequently built from a network of
cubic patches, usually NURBS. The control points defining these patches are manipulated to sculpt the
surfaces of the figure. Smooth continuity must be maintained at the edges of the patches, which often
proves challenging. Complex topologies also can cause difficulties, given the rectangular nature of the
patches. While patches can easily provide much smoother surfaces than polygons in general, it is more
challenging to add localized detail to a figure without adding a great deal of more global data. Hier-
archical splines provide a partial solution to this problem [ 21 ].
Other representations
Several other methods have been used for representing virtual human figures. However, they are used
more infrequently because of a lack of modeling tools or because of their computational complexity.
Implicit surfaces ( Chapter 12.1 ) can be employed as sculpting material for building virtual humans.
Frequently the term “metaballs” is used for spherical implicit surfaces. Metaballs resemble clay in their
ability to blend with other nearby primitives. While computationally expensive to render, they provide
an excellent organic look that is perfect for representing skin stretched over underlying tissue [ 35 ] [ 41 ].
Subdivision surfaces ( Chapter 12.3 ) combine the topological flexibility of polygonal objects with
the resultant smoothness of patch data. They transform a low-resolution polygonal model into a smooth
form by recursively subdividing the polygons as necessary [ 17 ][ 19 ] .
Probably the most computationally demanding representation method is volumetric modeling.
While all of the previously mentioned techniques merely store information about the surface qualities
of a virtual human, volumetric models store information about the entire interior space as well. Because
of its extreme computational requirements, this technique is limited almost exclusively to the medical
research domain, where knowledge of the interior of a virtual human is crucial.
As computers continue to become more powerful, more attempts are being made to more accurately
model the interior of humans to get more realistic results on the visible surfaces. There have been sev-
eral “layered” approaches, where some attempt has been made to model underlying bone and/or muscle
and its effect on the skin.
Chen and Zeltzer [ 14 ] use a finite element approach to accurately model a human knee, representing
each underlying muscle precisely, based on medical data. Several authors have attempted to create
visually reasonable muscles attached to bones and then generate skin over the top of the muscle
(e.g., [ 57 ][ 66 ]). Thalmann's lab takes the interesting hybrid approach of modeling muscles with meta-
balls, producing cross sections of these metaballs along the body's segments, and then lofting polygons
between the cross sections to produce the final surface geometry [ 10 ]. Chadwick et al. [ 13 ] use free
form deformations (FFDs) to produce artist-driven muscle bulging and skin folding, as described in
Section 9.1.5.
9.1.2 Geometry data acquisition
Geometric data can be acquired by a number of means. By far the most common method is to have an
artist create the figure using interactive software tools. The quality of the data obtained by these means
is of course completely dependent on the artist's skills and experience. Another method of obtaining