[ptx] Using nona before autopano-sift for ultra wide-angle lenses

[Daniel M. German]

 Sebastien Perez-Duarte twisted the bytes to say:

 Sebastien> Hi all
 Sebastien> I have a very wide angle lens (Canon EF-S 10-22mm) I use for
 Sebastien> equirectangular panoramas. One big problem I have is that
 Sebastien> autopano-sift is not very efficient with the corner of the images,
 Sebastien> where the deformations are greatest. The control points found are all
 Sebastien> in a small central area, hence stitching is not always good, and I
 Sebastien> have to manually add control points in the corners... and then, why do
 Sebastien> I use autopano-sift ?

I also use this lens, but I don't use autopano-sift. I usually set the
points by hand,

but my suggestion is to correct the lens first. Use clens for that. It
will remove the distortion (although the edges will still be naturally
distorted due to the fact it is a rectilinear lens).

Another suggestion is to crop the image before giving it to autopano.


 Sebastien> I would lilke your feedback and your help with a new workflow that (I
 Sebastien> think) would help autopano-sift.

now with respect to your idea:


 Sebastien> 1) I know how the images are shot, give or take a few degrees: one row
 Sebastien> at 30° down, one row at 30° up, one above and one below. I could then
 Sebastien> use nona or PTStitcher to transform these images in an equirectangular
 Sebastien> projection (-30° and +30°), using the simplest parameters.

this is easy. Create automatically a script for PTmender.

 Sebastien> 2) autopano-sift could then be used in these transformed images, where
 Sebastien> the deformations apparent in the original images have been erased for
 Sebastien> the overlapping images.

Then give the resulting images to autopano (perhaps in pairs, given
that you know the location of the points).

 Sebastien> 3) Some process (hopefully not very hard to program out of the PT
 Sebastien> tools) then projects the control points found on the transformed
 Sebastien> images back to the original rectilinear coordinates.

The .pto file is easy to generate from the ones you get form autopano.

 Sebastien> 4) hugin is run as usual with these control points and the original
 Sebastien> images. The rest of the usual workflow follows.

then run hugin.


 Sebastien> What do you think ? do you have an idea how the point 3) could be done ?

it sounds doable.



 Sebastien> Thanks in advance

 Sebastien> S

--
Daniel M. German 
http://turingmachine.org/
http://silvernegative.com/
dmg (at) uvic (dot) ca
replace (at) with @ and (dot) with .