If you aren't ready for quaternions (there's plenty of example source code on the internet), you can convert the matrices to axis-angle format, resulting in a vector and a scalar, which you can interpolate directly.

An even simpler method is to interpolate two rows or columns of the matrices as vectors, then create a new matrix from the resulting two interpolated vectors: cross (product) the two vectors to get the missing third vector, then orhonormalize (normalize/cross or dot-project (Gram-Schmidt)) the vectors to create an orthogonal matrix (3 orthonormal vectors make up a rotation matrix).

The first method may produce better results, and quaternions give more options for interpolation (lerp (linear), slerp (spherical linear), squad (cubic)).

Try www.google.com for more info on any keywords in this post.

Quote:Original post by jyk
Here's another option to consider. You can get the same interpolation as you would with quaternion slerp (albeit at greater expense) by finding the matrix that rotates from A to B, extracting the axis and angle from this matrix, scaling the angle, and then recomposing the matrix from the axis and angle and multiplying with A.

I'm at work now, but if you're interested in this method I can post details later (or perhaps someone else will in the meantime).

Linear interpolation (Vectors, scalars):
delta = b - a;
// alpha = [0..1]
c = a + delta*alpha

3x3 rotation matrix form (right-hand element evaluated first):

delta = b * transpose(a) // transpose(a) followed by b.
delta.getAxisAngle(axis,deltaAngle)
// alpha = [0..1]
c = axisAngleToMatrix(axis,deltaAngle*alpha) * a