My 2 cents
Why don't we start using focal length rather than FOV as the basic angular scale parameter? It is a pure scale factor, independent of image size and the shape of the projection function, both of which are conflated in FOV. In my experience, 3/4 of the confusion about image scaling and angle-of-view calculations comes from the difficulty of disentangling those 3 basic parameters from FOV.
I believe that this needs to be moved to panotools-devel or hugin-ptx.
Is everybody ok with this? I will then repost on these lists.
D M German wrote:
One of problems I encountered when trying to do very long mosaics of a
wall, is that I had to "fake" the FOV of each photo to be very small.
Also, another problem of libpano is that horizontal/vertical shift is
done before the image is mapped to the panosphere (and before the lens
correction is done). But for plannar mosaics you want to do the
translation after the image has been mapped to the infinite plane.
Actually, this shift is a part of the lens correction and not designed for a planar mode. This is why an additional shift at the right place would be much better.
Pablo> They are the normal FOV of the camera that was used for capturing the
Pablo> image. This also means that all the distortion correction mechanisms
Pablo> etc. still work properly, and one can reuse the lens calibration for
Pablo> all images that were captured with the same camera (and optimize them
This was the main motivation of the plannar mode.
Then I don't see why it is necessary to make such intrusive changes.
If the panorama projection is set to rectilinear, the panorama will be identical to the image on the plane, and one basically gets a planar mode for free. It is not very useful with the default r,p,y parameters, though. This is why I added the Tr ones.
I just saw that the optimizer always minimizes distances on the sphere and not in the output image. While this might makes sense for the panoramic usecase, it is probably not desirable for creating a planar mosaic.
> One does not have to
fake extremely narrow FOV for the lens to fit the mosaic in a sphere
(although, from what you say Pablo, the image will just keep wrapping
around the sphere several times).
No, that is not possible with a rectilinear panorama anyway.
> By having a plannar mode we can use
the proper FOV of the lens.
Actually, this is also possible with the Tr parameters.
However, I found that with the additional parameters, optimisation generally tends to become more unstable, and I'm not sure if the Tr or a modified Ti (as used by PTGui and PTStitcherNG) is better in that respect.
The Planet: dedicated and managed hosting, cloud storage, colocation
Stay online with enterprise data centers and the best network in the business
Choose flexible plans and management services without long-term contracts
Personal 24x7 support from experience hosting pros just a phone call away.
PanoTools-devel mailing list