Started by Feb 16 2005 05:32 PM

,
6 replies to this topic

Posted 16 February 2005 - 05:32 PM

Hey everybody. I've been working on my engine, and I figure it's time to get some stuff moving in it. I'm about to start working on my skeletal animation code, and I was wondering if I understood certain concepts correctly. I'm working based on the article on real soon now.
I'm a little confused as to how to use the relative and absolute matrices of of the joints. Here's how I see it in my head. A joint will have an absolute, relative, and final matrix. The absolute will hold the pos/orientation of the joint. The relative matrix will hold the pos/orientation relative to the parent's pos/orientation. The final matrix will hold the current pos/orientation at a point in the animation. The final matrix is calculated by post multiplying your parent's matrix by your relative matrix (the offset from the parent) by the joint's current animation matrix, i.e.,
final = ParentFinal * MyRelative * MyFinal;
Am I missing anything? Is the absolute matrix in there anywhere? Am I multiplying in the right order. I'm just trying to get things straight in my head before I started coding a couple of hours worth of bugs :) Thanx for the help.

Posted 17 February 2005 - 10:51 AM

hmm... is my description unclear or has no one done skinning code around here?

Posted 17 February 2005 - 12:12 PM

I think your terminology is a little unusual...I don't understand the difference between what you call "final" and "absolute".

In any case, I would compute the matrix that you use to actually multiply a vertex as:

ModelToWorld = InvBindPose * MyRelative * (concatenation of all parent's relative matrices)

The bind pose matrix describes the transformation of a joint in it's "resting" or bind position. The vertices are initially specified in the model's bind pose (kind of like "root model space"). The inverse bind pose transformation moves vertices for a particular joint into the joint space, and then all the relative matrices move the vertex back into world space, but using the current animation transforms. If you aren't familiar with this idea already, you should probably make sure you have a good handle on this first, certainly before coding anything. There is information on the net about this.

The relative matrix is the transformation from the child space to the parent space. This one changes every frame according to your animation.

This also assumes that the ultimate root node contains a transformation of the entire model to world space.

As for the order, whether it is written R to L or L to R depends somewhat on convention; the way I write it above, I expect InvBindPose to be applied "first".

Usually what will happen is that you will compute all of the matrices for each bone, and then you'll compute a weighted average of the vertices to do the actual skinning. Perhaps you are already familiar with that part.

In any case, I would compute the matrix that you use to actually multiply a vertex as:

ModelToWorld = InvBindPose * MyRelative * (concatenation of all parent's relative matrices)

The bind pose matrix describes the transformation of a joint in it's "resting" or bind position. The vertices are initially specified in the model's bind pose (kind of like "root model space"). The inverse bind pose transformation moves vertices for a particular joint into the joint space, and then all the relative matrices move the vertex back into world space, but using the current animation transforms. If you aren't familiar with this idea already, you should probably make sure you have a good handle on this first, certainly before coding anything. There is information on the net about this.

The relative matrix is the transformation from the child space to the parent space. This one changes every frame according to your animation.

This also assumes that the ultimate root node contains a transformation of the entire model to world space.

As for the order, whether it is written R to L or L to R depends somewhat on convention; the way I write it above, I expect InvBindPose to be applied "first".

Usually what will happen is that you will compute all of the matrices for each bone, and then you'll compute a weighted average of the vertices to do the actual skinning. Perhaps you are already familiar with that part.

Posted 17 February 2005 - 05:56 PM

Ah ok, makes sense. One question though. Is the InvBindPos mat relative to the world origin or relative to your parent bone?

Posted 17 February 2005 - 06:43 PM

Greetings!

Don't mess arround with too many matrices. You need the following matrices:

Local matrix - rotation & position for local joint

World matrix - world = local * parent->world (this will ensure that you get the chain)

Inverse world matrix - you need it to transofrm vertices through it

Animation world - joint in world space

Vec3TransformCoord(&vNew, &inverseworld, &OriginalVertex); //delta vertex

Vec3TransformCoord(&vNew, &AnimationWorld, &vecstart); //vertex in world space ready to render

Don't mess arround with too many matrices. You need the following matrices:

Local matrix - rotation & position for local joint

World matrix - world = local * parent->world (this will ensure that you get the chain)

Inverse world matrix - you need it to transofrm vertices through it

Animation world - joint in world space

Vec3TransformCoord(&vNew, &inverseworld, &OriginalVertex); //delta vertex

Vec3TransformCoord(&vNew, &AnimationWorld, &vecstart); //vertex in world space ready to render

Posted 18 February 2005 - 06:00 AM

The inverse bind pose is the matrix that moves a point from model space to bone space (each bone has it's own). So I guess you would say it's relative to the model root space. This matrix should come out of your modeling program, one way or another, and it's a constant for each bone in a model.