I've been experimenting with skin over an IK-bones model:
In this test the skin is a mesh of bicubic patches (forming a cylinder), and
I have transformation matrices for the upper (Ma) and the lower arm (Mb).
The control points belonging to the upper arm are transformed only with
matrix Ma, those belonging to the lower arm only with matrix Mb. The three
control_point sets near the joint are weighted this way:
wa = 0.75    wb = 0.25
wa = 0.50    wb = 0.50
wa = 0.25    wb = 0.75
and the control points are transformed to

V' = V * Ma * wa + v * Mv * wb

The rotation of the control points near the joint looks good, but somhow
they appear 'scaled down', causing the skin mesh to get very tight and thin
in the joint-area.
I was toooo frightned to attach a picture (although it's only 3k) in a
posting to this list ;-)
It can be seen here: http://192.94.226.61/arm.jpg
It shows the IK-bones and the transperent skin with control-points when the
arm is bended by 90 degrees, and it gets even worse on higher angles.

Now my quesion is
- is this the normal behavior of multimatrix vertex blending or is it a
'bug' in my sdl-implementation ?
- if it's not a 'bug', what can be done to avoid it?

Thanks a lot!
-sascha

Post a reply to this message