Linear Panorama with Fisheye Lens

[jürgen eidt]

Hi everyone,

I’m working on a linear panorama and ran into an issue when switching from a wide-angle lens to a fisheye lens.

With the wide-angle setup, everything was straightforward: I placed horizontal and vertical control points within each image, and horizontal control points between images. Using rectilinear projection, the optimization worked well and produced a clean result with a narrow field of view.

Now I’m trying the same approach with fisheye images. The idea is to reduce the number of shots and resolution requirements. However, the panorama doesn’t come out straight. Each image still shows strong barrel distortion, even though I’ve added horizontal control points.

When I first convert the fisheye images to rectilinear and then stitch them, it works fine. But I’m wondering:

• Is there a limitation when using fisheye projection directly for linear panoramas?
• Is this workflow even supported, or am I missing something?

For reference, here is the project file with the images:
https://bitfabrik.io/tmp/Mural-1x6.zip (66MB)

Thanks
Jürgen

[Erik Krause]

Am 21.08.2025 um 19:44 schrieb jürgen eidt:

> I’m working on a linear panorama and ran into an issue when switching from
> a wide-angle lens to a fisheye lens.

[...]

> Now I’m trying the same approach with fisheye images. The idea is to reduce
> the number of shots and resolution requirements. However, the panorama
> doesn’t come out straight. Each image still shows strong barrel distortion

I assume by "linear panorama" you mean that type, where you move the
camera parallel to a mural, a street front or similar. The big problem
of those panoramas is parallax for anything in front of or behind the
plane of the mural. And parallax gets worse, the more sideways the lens
looks. Hence, I don't think using a fisheye is such a good idea.

However, that is not the source of your problem. For a standard single
viewpoint panorama, PTGui uses information from the overlap area to
determine focal length and lens distortion. In case of a linear
panorama, this is not possible.

So to have this working with a fisheye lens, you would need to use fixed
lens correction data, that is focal length, fisheye factor and lens
correction data. If you use that data from a standard, parallax-free
spherical panorama shot with the same lens, it should be possible to
avoid the barrel distortion.

Easiest is to load a suitable spherical panorama project, save the lens
data into the lens database (Lens Settings tab) and load them from the
database in your linear panorama. Then disable optimization of all lens
parameters (Optimizer tab, advance interface), that is deselect anything
in the "Global lens profile" row and don't use any individual parameters
for images (deselect on Lens Settings tab).

--
Erik Krause
http://www.erik-krause.de

[jürgen eidt]

Thanks Erik for your suggestion!
This linear panorama is a simple mural that is flat and has no objects in front of it, so I thought this would be a good case for the fisheye lens. Compared to the wide-angle setup that required 11 shots, the fisheye allowed me to capture the scene with just 6 images.
I manually configured the lens parameters and experimented with various settings to see if there was a path that could yield better results, but unfortunately, nothing promising came out of it.
For example, setting the focal length to large value of 500mm looks something that could work, but the individiual images still have the strong barrel distortion. Correcting 'b' manually does not seem like the right path either.

Maybe I'm missing something?

Jürgen

[Erik Krause]

Am 21.08.2025 um 21:40 schrieb jürgen eidt:

> For example, setting the focal length to large value of 500mm looks
> something that could work, but the individiual images still have the strong
> barrel distortion. Correcting 'b' manually does not seem like the right
> path either.
> Maybe I'm missing something?

I missed something. Obviously, the very-long-focal-length trick only
works with rectilinear images. The aim of this trick is to simulate an
orthographic projection that would resemble a viewpoint at infinite
distance. Such a viewpoint would not allow for fisheye projection, it is
always rectilinear.

It should also be possible to use viewpoint correction for linear
panoramas, in which case you could use the true focal length and
characteristics of the lens. That would be the only possibility to
actually use fisheye images. I never tried this, so there might be limits.

[Erik Krause]

Am 21.08.2025 um 22:54 schrieb 'Erik Krause' via PTGui Support:
> It should also be possible to use viewpoint correction for linear
> panoramas, in which case you could use the true focal length and
> characteristics of the lens. That would be the only possibility to
> actually use fisheye images. I never tried this, so there might be limits.

I tried that on your images (should have done it in the first place).
While it should work in theory, it's way outside the intended use, and
it needs a lot of manual work, especially setting control points. I
attach the project file.

[jürgen eidt]

Obviously, the very-long-focal-length trick only
works with rectilinear images. The aim of this trick is to simulate an
orthographic projection that would resemble a viewpoint at infinite
distance. Such a viewpoint would not allow for fisheye projection, it is
always rectilinear.

Thanks for the explanation. It makes sense, and I guess that was what I was missing. It is like trying to make a curved mirror reflect straight lines.
I never thought of using viewpoint correction for this, but it is a clever and creative use of it.
Also, thanks for sharing your .pts file. Much appreciated!

Jürgen