[ptx] autopano: first attempt at semi-automatic panorama stitching

[Ed Wildgoose]

>Looks very nice. I'm currently hacking together a simple program that uses
>the phase correlation mentioned below to create initialisations for the KLT
>code above, which already includes a feature detector. Hopefully this will
>work for simple rectilinear panoramas, without much rotation.
>
>First experiments looks like it'll work for single row panoramas :)
>  
>
My *guess* is that this will work extremely well for most panoramas with 
28mm lenses and above, but do you think that it will work well for 8mm 
fish eye scenes, or even just very wide angle lenses?

I wonder if you could *assume* something like a grid layout and 
pre-process each image pair based on expected geometry and then apply 
phase correlation.  eg if you expect the images to be either left or 
right aligned then coursely correct for an assumed 30% overlap, and go 
with that.  At least that way the images are roughly capable of being 
lined up?

Along the same vein, perhaps there is some mileage in subdividing an 
image up into thirds down each axis, precorrecting based on an assumed 
lens focal length and then phase correlate all 9 pieces against 
everything else...  The intention would be to catch corner matches and 
the like..

Just a wild idea.  Sorry if it's noise.