[sldev] Anyone here with OpenCV experience?

[Tigro Spottystripes]

from what i remember of an old version of that program, you can change
the positions it expects to find the dots in 3d as long as you follow a
simple rule (like for 4 dots  they can't be coplanar, I'm not sure about
the requirements for 3 dots), so a generic pattern for the position of
things in the face should work with minimum calibration


Jan Ciger escreveu:
> Tigro Spottystripes wrote:
> > I'm talking about using the 6dof calculation algorithm, but inputing 3
> > or 4 dots that were acquired a different way, like say, the eyes and the
> > nose identified by another algorithm
>
> Yes? You probably do not realize that those points need to be in a
> specific spatial relationship. The original algorithm works by fitting a
> model of the rig with known size into the 2D image. If you change the
> size (everyone's face is different), it won't work anymore.So you cannot
> just feed arbitrary 4 points acquired from a different place.
>
> Otherwise you would need at least two cameras in a calibrated rig (like
> that Minoru webcam) to be able to track anything more that 2DOF.
>
> Regards,
>
> Jan