On Tue, Nov 22, 2011 at 11:46:01AM +0000, Bruno Postle wrote:
> On 22 Nov 2011 08:15, "Rogier Wolff" <[email protected]> wrote:
> >
> > I thought that for a horizontal controlpoint pair, the lattitude
> > simply doesn't count. So all that the optimization step cares about
> > is the that they line up horizontally.
> >
> > Similarly for the vertical control lines. There the horizontal position,
> > or longitude is not taken into account.
>
> This is effectively true with equirectangular, or any of the other
> cylindrical output projections.
>
> > I thought that all this was independent of the projection
> > being used for the final result.
>
> Nope, the horizontal and vertical points are evaluated in the output
> canvas, so the output projection is critical.
>
> This is much simpler conceptually, as far as the optimiser is concerned
> they are the same thing.
Do you mean that the control point matching happens in ouput
projection space?
i.e. a controlpoint in Image1 at X1, Y1, and in Image 2 at X2,Y2 is
transformed using the parameters for Image1 (i.e. roll1, pitch1) to a
roll/pitch coordinate pair in the pano-sphere and then onto the output
canvas using the output transformation?
The same is then done for the X2, Y2, and the difference is optimized.
This would mean that for instance a mercator projection that has a
distortion near the poles will favor the lattitude of controlpoints
near the pole being "perfect" sacrificing all other overlaps.
It would also mean that after changing the output projection, you need
to optimize again.
A friend shoots "all around" panoramas. As output projection she needs
a projection onto a cube around the pano-sphere. So if I understand
things correctly, she will set the output projection to
equirectangular, stitch an output image, rotate the viewpoint by 90
degrees and stitch another face of the cube until all 6 faces are
done.
Now optimization is probably done with the output projection set to
one of the faces. Now all control points that lie outside the face of
the cube are distorted and optimized in weird ways that do not reflect
their role in the final output.
Of course something can be said for doing it this way: if there is a
minute difference in the projection of the layout of two images near
the center of the output image, and the same minute difference in
degrees on the panosphere expands to several tens of pixels near the
edge of the output image, it might be good to "fix" that controlpoint
near the edge, and tolerate a slightly larger error on the one in the
middle.
But I would prefer to optimize in panosphere coordinates. Doing it the
other way introduces errors based on the assumption that all
controlpoints are perfect. They are not. And the lens parameters are
not perfect.
I think we'd get a much better fit (in a mathematical sense, on the
panosphere) if we'd just use the panoshpere coordinates.... Once we
have that, we're ready to optimize lens parameters etc etc, to get the
final errors out. And then a "list the controlpoints starting with the
largest error" allows you to find the controlpoints that really have
errors in placement.....
But again... It's entirely possible that I'm misunderstanding how
hugin actually works. (I'm reading "cooking for geeks" and the book
has explained one simple thing to me and something I didn't manage
before (again and again) worked first time, simply because now I
understand the underlying chemistry. Similarly I want to know how
hugin works to be able to better control it.)
Roger.
--
** [email protected] ** http://www.BitWizard.nl/ ** +31-15-2600998 **
** Delftechpark 26 2628 XH Delft, The Netherlands. KVK: 27239233 **
*-- BitWizard writes Linux device drivers for any device you may have! --*
The plan was simple, like my brother-in-law Phil. But unlike
Phil, this plan just might work.
--
You received this message because you are subscribed to the Google Groups
"Hugin and other free panoramic software" group.
A list of frequently asked questions is available at:
http://wiki.panotools.org/Hugin_FAQ
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at http://groups.google.com/group/hugin-ptx
