Turning StickFigure into FatAss:
Skeletal Skinning
by Thomas 'ThamasTah' van Dijk

Please note that it is recommended that you read my Skeletal Modelling tutorial.

Introduction

Whatever you build using my previous version of Skeletal Modelling isn't going to be overly realistic; just lines. This got me thinking. I needed a way to define the 'thickness' of a bone. Just increasing the DrawWidth isn't going to be very convincing ;), so a totally different system was needed.

Theory

This is actually quite like Skeletal Modelling...

We take the middle of a bone and from there define a bunch of vectors. The angle of the vector will again be relative to the bone, but this time, also the length is relative.
An illustration:

(Okay, it looks crap, but you get the idea, don't ya? :))

The big fat vertor is the bone, the little vectors are the surface descriptors (originating from the centre of the bone), the shaded area is what we're trying to define.
If we change the angle of the bone, the descripors turn along; if we change the length of the bone, the descriptors scale along.
Note that we do need a way seperate the two sides...

We need something else, though. If you added a bone, you'd get the following:

That doesn't look look like an arm at all. Sticking to the current system, we could solve it:

But this severly screws up when we bend it:

No good, indeed. You can actually see this in early 3D games, most noticable example being Tomb Raider (heh, an excuse for doing some more 'research' on Lara... ;D Seriously though!) Check her elbows and knees. This was fixed in version 4, but, ihmo, in an inferior way.

What we do, is use the angle bisector. On both sides we define a length (which probably should be relative to the average of the length of both bones). We add two points (one at each side of the connection) on the angle bisector with the defined distance to the connection, e.g.:

We connect the surface descriptors not through the connection point, but through those points:

Mo' betta stuff, ain't it?

Conclusion

Again, this could 'easily' be adapted to 3D, where it full potential shows.

I'm currently trying to construct an algorithm for doing collision detection on lines. Is it actually simple and is it just me, or are you folks also having trouble with it? Cya!

Thomas 'ThamasTah' van Dijk
paxdev@hotmail.com
dec 4 2000