Skinning and Animations - WebGL Game Development - page 291

// flatten bone matrices to array
for ( var b = 0, bl = this.bones.length; b < bl; b ++ ) {
// compute the offset between the current and the original
transform
// was already representing the offset; however, this
requires some
// major changes to the animation system
mat4.mul(offsetMatrix, this.bones[ b ].skinMatrix,
this.boneInverses[ b ] );
this.flattenToArrayOffset(offsetMatrix, this.boneMatrices,
b * 16 );
}
};
}();

In the preceding code, UpdateWorldMatrix is the key function where all the
computations happen. It performs the following steps:

1. It first computes its own modelMatrix and then updates its matrixWorld
matrix (concatenates parentMatrix ).

2. It iterates over the children array. If the child object is an instance of bone ,

then it invokes the update function to compute its skinMatrix . If the child
object is an instance of StageObject , it invokes updateWorldMatrix of the
child object to compute its matrixWorld matrix.

It then computes the offset matrix for each bone in the array. When the object of
RiggedMesh is first initialized and updateWorldMatrix is invoked, it checks whether
the boneInverses array is created. If they don't exist, it computes the inverse of
each bone's skinMatrix and stores it in the boneInverses array. The bone's initial
matrix is the binding matrix. This initial call stores the inverse binding matrix. Then in
subsequent calls of updateWorldMatrix , it does not calculate the boneInverses array
(discussed in the The binding matrix section of this chapter). The inverse binding matrix
is only calculated once. It then computes offsetMatrix i =B -1 i *skinMatrix . While
animating bones, skinMatrix will change each time, but the initial inverse binding
matrix is calculated only once and stored in the boneInverse variable. When we
load the 08-Loading-Skinned-Models.html page in a browser, the character model
maintains its starting pose. This happens because the initial skin matrix is multiplied
by its own inverse. Hence the offset matrix for each bone becomes an identity matrix,
so no vertex is transformed. During animation, a vertex is transformed from the skin
local space to joint local space, and then we transform it using the
joint's skinMatrix .