M1 =3D M2 * M3 only if M1 and M2 are perfectly correct. There are =20
always errors, unless you have perfect camera calibration and perfect =20=
subpixel corner finding for the four corners of the markers; neither =20=
of these will be the case with any vision system! So, small errors =20
in both these will result in small (or even large) errors in the pose =20=
computation of the camera relative to the marker. Thus, M1 and M2 =20
will have error, and those error will be magnified significantly when =20=
you invert one of the matrices (since a small amount of rotation =20
error on the marker will result in huge translation errors in the =20
A better approach would be to figure out the fixed relationship of =20
the 2 cameras on the rig, get the four corners of the markers in the =20
two images, and use a pose computation that takes all 8 corners =20
across both cameras into account! The stereo info should give you a =20
huge improvement (shouldn't it, folks?).
I thought that people had done this already with the AR Toolkit ...
But, if your goal is to use the ARToolkit to find the relationship =20
between the cameras, it's not accurate enough. I would suggest =20
looking on the web, there have to be tools for doing this; I know =20
folks do this sort of thing all the time (multicamera calibration).
On Jul 9, 2005, at 8:52 AM, joele.d@p ........ wrote:
> Hello everybody,
> We have 2 camers mounted on a rig. We can get the transformation
> matrix for the first camera and for the second.
> Now we are trying to discover a relation between the 2 matrices. Does
> anybody know that relation?
> I have been thinking. Basicly you have one image of the marker in the
> first camera,
> and a second image of the (same) marker in the second camera. Could I
> use, in same way, the RelationTest example? I thought I could use the
> relation that the inverse of the first matrix multiplied by the second
> matrix gives the relation between marker2 and marker1,let=B4s say M3.
> Would that mean the following: M1 =3D M2*M3?
> Then we are going to move the rig, and the marker stays fixed. Then,
> does that relation still hold? I think yes, because with respect to
> each other the cameras do not change.
> Any comments are welcome, please.