new features (estimate ala PanoPoints)

[Pablo d'Angelo]

On Mon, 27 Oct 2003, Peter Suetterlin wrote:

> Pablo d'Angelo wrote:
> > Hi,
> > 
> > I've added another feature to the control point picker, inspired from
> > PanoPoints. It is possible to estimate the translation of the images by
> > averaging the control points.
> 
> Hmm, it shouldn't be too difficult to get translation *and* rotation.
> maybe use the normal optimizer functions?  For only a few points that
> should be very fast.  Then the matching should be better even with
> non-horizontal images - maybe good enough to feed that position to the
> auto picker.

Yes, thats what I planned from the start. but well, since I don't have an
infinite amount of time, it was easier for me to write a simple loop
to average the translation than to write an interface for the libpano12
optimize functionality. The advantage of the optmizer approach is that it
would be that it would work for fisheye images and all the other image types
supported by panotools as well.

> Further, after having determined rotation and shift - you think it's
> possible to (roughly) determine the common area, look for
> high-contrast features there and auto-map them to the second image?

Thats definately possible. actually I'd like
to try and implement a global approach first (like the one in the
"Recognizing Panoramas" paper of Brown), sicne the ultimate goal is to get
rid of that point picking at all :)

ciao
  Pablo
--
http://wurm.wohnheim.uni-ulm.de/~redman/
Please use PGP