Re: [Algorithms] skeletal animation vs. morphing

[Michael D. <md...@io...>]

----- Original Message -----
From: "Eric Maslowski" <mas...@ms...>


> Hello,
>     I was wondering if anyone could foresee any faults in using a morph
based
> animation system over a skeletal animation system for character animation?
By
> "morph based" I mean a system that is defined by poses. These poses are
created
> by the artist (possibly with help from a utility that speeds up common
poses).
> Which pose to use is decided at run-time depending on in-game criteria.
Simple
> linear interpolation would be used to blend from one pose to the next.
It's your
> standard morph-target animation system.
>     It seems to me that it would be very efficient at run-time. May take a
little
> longer to create content for, but the resultant animation can be much more
> dynamic. Any thoughts? Thanks.

This approach is certainly workable, and it is the animation system that
Quake used up through Quake 3.  The problem with morph target animation
systems are in part:

1) Large storage requirements.  The higher the poly count in the model, the
worse the storage requirements become.  Also adding new frames of animation
is expensive in terms of memory storage.

2) Linear interpolation.  This means vertices move directly from one
position to another, and thus don't handle curves well.  All of the body's
movements are along curves, so you need more keyframes to approximate the
curve to keep it smooth, which just makes storing the animation even more
memory prohibitive.  The fewer keyframes you have, the more likely the
animation will be distorted.

3) Difficulty in blending.  If you are wanting to blend two separate
animations together morph target animation systems don't work as well as
skeletal animation systems.  The keyframes in your morph target animation
were meant to flow into each other, so you can check during content creation
to avoid bad distortion.  But when you blend different animations together,
the chance of getting some messed up distortions are higher.  With skeletal
animation systems you can blend as many keyframes together as you like, and
you won't get mesh distortion.  You may get a screwed up pose, but you won't
get mesh distortion.

4) Animation locked to mesh.  When you work with a morph based system, the
animation is "baked" into the mesh.  With skeletal systems, you can attach
different similarly sized meshes to the same skeletal system and re-use the
animations quite easily.



That said, morph target based animation systems are a LOT easier to
implement, and you can use existing model formats like the Quake formats to
store and load your models (if you don't plan on selling your engine).  And
as you said, it is faster to calculate frames for a morph based animation
system than for a skeletal based one.  And it is easier to export morph
targets from an animation package than skeletal system data.  The morph
targets can just be exported as separate models, but the skeletal system
requires exporting a heirarchical skeletal system, exporting the animations
for the skeleton, and exporitng meshes with the correct weighting
information to attach the vertexes to the various bones of the skeleton.


Hope this helps,
Michael Duffy
md...@io...