Tuesday, February 07, 2006

To hack or not to hack?

I'm wondering if I'm the only person in the whole world who uses anim8or to generate test images with known camera parameters and then uses MATLAB to work with the images, with those known camera parameters...

Hmm... I guess I'm not the only one as I know SQ is using it as well. (we sit back to back in the lab.)

There's some weird thing I don't understand why. The images from Anim8or have a half pixel distance difference from what is calculated from Matlab?! For example, a textured cube with 8 vertices in space, I know the vertices, I know the position and pyr of the camera, so in anim8or, I position everything as desired and render the image. No problem! When I get this image into Matlab, I calculate the image points using the vertices and camera matrix and overlay them on the image, strangely enough, the two set sof points don't exactly match. But, it matches exactly when I add 0.5 to all the image points I calculated in Matlab!

So, to have a good morph (as that's what my research is on in case you don't know), I have to have the feature points exactly on. But it'll screw up the fundamental matrix and the projective geomatry because the image point is HACKED!

Well... What should I do to get over this? The great spirit of the universe, please enlighten me!

No comments: