template matching (semi-automatic finding of control points)

[Bruno Postle]

On Mon 12-May-2003 at 08:18:46PM +0200, Pablo d'Angelo wrote:
> 
> I just checked it into the current cvs, so that people can play
> with it :) and give some feedback.

I built RPM packages of it and various other stuff (including that
wxWindows panoviewer):

    http://bugbear.blackfish.org.uk/~bruno/panorama-tools/

The i386 rpms are all built for redhat-9; but there is nothing
special going-on, so the source rpms should rebuild on any
recent-ish linux.

On Tue 13-May-2003 at 08:20:03AM +0200, Pablo d'Angelo wrote:
> > 
> > I've had a bit of a play with the auto control point selection.
> > It works very well;
> 
> How fast is it on your machine with bigger pictures?

It's _really_ slow with big rectangles, though this is on an old
celeron500.  Small rectangles (30x30-ish) with 1280x960 images are
reasonably fast and it feels easier than fine-tuning by hand.

> > This _could_ be very automated, even for fisheye sources; the
> > user would first have to roughly position the images though,
> > since trying all possible rotations &c might get very slow.
> 
> Small differences (+-5 deg or so) are more or less tolerated.  You
> are right, if the roll angle is known from somewhere it could be
> incorporated (by simply rotating the template).

If the template image is remapped (using panorama tools and
estimated parameters) to roughly the same projection as the other
image, it could be aligned very accurately (in principle).

> but usually the rotation is very small anyway, except for very bad
> handheld shots.

That would be me and my photos ;-)

-- 
Bruno