Fiseye lens distortion

[Vizioen]

Hi all,

I'm in need of some assistance. I need to match a 3D model into a fisheye lens photo provided by the client. As I have never done anything like this I was already testing how I would tackle this problem. And I was looking into the quadratic/cubic distortion modes. How does this relate to the real life fisheye lens distortion? Or should I leave this and work with the fish eye lens settings in the render settings. Also can these values be matched to the real life fish eye lens distortion if you know the make/type, focal distance?

Also auto fit in the fisheye lens distortion in settings, doesn't work with vray RT

[tzaar]

Luckily I have never encountered this (nightmare) more than as an experiment of thought!
Do you have any info about what lens and focal length was used?
My first idea, depending on how good quality/resolution the photo is, and how bad distorion is. I would try maybe correcting it to linear with Photoshop, Lightroom, PTLens or something like that. And then match and render as usual and then, distort things back to fisheye in Photoshop. You might loose some pixels in the edges, and definitely loose quality this way. But it is one way... if all other fails.

[joconnell]

One of the issues with lenses is that they aren't always totally even - some fish eyes have a slightly different distortion amount at the edges than they do at the front. A good idea is if the client has the lens itself, get them to shoot a grid on a monitor from as square on as possible with the lines of the grid running out of the edges of frame - then you'll have some reference lines to see how the distortion works. Undistort the image, do your perspective match and 3d work, then redistort the 3d to match the original photo.

[Syclone1]

http://www.photozone.de often has grid shot of different lenses already (usually under analysis heading in the lens reviews), if you can find out what camera body and lens was used then you might be able to somewhat match it with something on there as a reference.

[Vizioen]

Thanks guys for the awesome ideas. The client is not in direct contact with me, I think there are like 5 different guys and 3 different companies between me and the client so asking for something more than, with what camera and lens has this been made, is not feasable. So that website with the grid shots will be a life saver if the lens they use will be up there. I still need to wait for the actual photo but I was just testing on how to distort with vray since I never did that. In what way does the quadratic/cubic distortion differ from the fisheye lens distortion?

[joconnell]

Quadratic / cubic was built to match the same parameters in the syntheyes lens distortion tool as far as I know. Fingers crossed you can get some jpegs from the client or the original raws and read the exif info in the file to get your camera and lens.

[Dave_Wortley]

If you've got video shot from the same lens you can run it through syntheyes and get your lens profile out from a motion track.

[Vizioen]

Ok so I got the photos and the information but I have no idea how to start on this I already tried a bunch of things but I think the easiest way is to distort it to rectilinear and distort it back. But I have no idea how to distort it back to fisheye. Following example is with and without distortion, I did this with the lens correction tool in photoshop but there's no way I can get from undistorted to distorted.





PTLens doesn't even succeed in undistorting it because he can't read the lens data in the exif. (There's none) While photoshop did it perfectly and it could read the lens data anyway (weird).

Camera model: Nikon D5200
Lens: Sigma 4.5 F2.8 EX DC

Even if I know how to distort the image back I have no idea what camera settings I should use with the undistorted image. Trial and error wouldn't be precise enough. I need this for a solar study so needs to be really precise. The contours of the building need to be more or less exact so it can be projected on a grid. The lens has an equisolid distortion. I would do it manually but no idea how to do it.

[Neilg]

You'd be better off using google street view and making low poly buildings. that's not a great photo to do a sun study with.