I am hoping that I can get some advice to solve a problem I am having with PTGui on a particular project.
I though I had my methodology honed – I use a Nikon D60 with a Nikon 10.5mm fisheye and take 8 horizontal images at 45° plus 2 zenith images at 90° and when I import these into PTGui and use “Align images..” the Optimisation usually reports “very good” with an “Average control point distance:” between 1 and 2 and occasionally this drops to “good” with an “Average control point distance:” between 2 and 3 if there is little detail for PTGui to generate the Control Points from.
When an important project came along a Nikon D300 with a Nikon 10.5mm fisheye was hired for the day because the Nikon D60, being an “entry level” DSLR, does not have a bracketing function and we wished to bracket 3stops either way (7 images per shot) to experiment with HDR.
Yes, I could have changed the exposure settings manually for each shot, but this would have been a real pain for one panorama, let alone 20.
The Nikon D300 was set up on the Nodal Ninja 3 so that the principal ray of the lens and axis of rotation were in the same plane and the “gold ring” aligned with the centre of rotation of the sphere when the NN3 was rotated, in exactly in the same way as I has set up my Nikon D60.
It was vital that this was done as accurately as possible because the centre of rotation of the camera and lens had to coincide with the centre of the HDS scanner we were using to a fraction of a millimetre, so a lot of care was taken.
When the Nikon D60 images are imported into PTGui they are all in portrait orientation, but the Images from the Nikon D300 came in with the 8 horizontal images in portrait and the 2 zenith images in landscape, which PTGui did not like, so this was sorted first, so according to my previous experience, and tests and panoramas I have made since, “Align images..” should have resulted in a near perfect result, but no.
The problem is that the area of overlap between to two sets of images is timber and painted brick so PTGui is unable to create Control Points between the images in the two sets although it created a mass of Control Points for the 8 horizontal images and a mass of Control Points for the 2 zenith images. This results in the zenith images appearing in the centre of the Equirectangular image, so I added lots of additional Control Point manually (including line CPs) for the overlaps of the zenith images with the horizontal images, but the result is horrible with huge distances between Control Points and “very bad” with “Average control point distance:” in the region of 120 and detail appearing twice some distance apart. This also results in the “Focal length:” of the lens being reduced (sometimes quite dramatically – e.g. to 8mm!) in the “Lens Settings”.
Using a Template from the Nikon D60 with a Nikon 10.5mm fisheye helps a bit, but not a lot.
There is no possibility of being able to retake the photographs or access to the Nikon D300 and Nikon 10.5mm lens used.
Any tips or suggestions as to how I can resolve this will be gratefully received.
sit to think about your problem and I really feel scarred. When I understand you right, you do not have overlap. I checked my databases, and I am afraid to say that the only reason I can imaging right now is that you did not change the lower rail setting from D60 to D300. With my personal NN hardware the LRS for the D300S is 61, for D60 it should be around 50. During NPP Tests I did and changing from one camera to another I sometimes missed to change the LRS and lost the pictures.
Use blended and layers to check if there is an overlay. When you used the wrong LRS I am afraid that you have a lot of parallax.
when you used the LRS of the D60 you have an offset of ca 10mm to the one of the D300.
In Ptgui go to advanced, image parameters, F1 should show "help for this tab". With mac use "Help" , "help for this tab". I have not tried this to correct a wrong LRS, but may be you can correct the viewpoint of your camera by changing 0 to 10 in VPx to correct the offset of your lower rail setting from 50 to 60. Might be this works. Otherwise try viewpoint correction from image to image. Just use the set of the best EV without the rest of the bracketing set.
I never tried to repair a lower rail offset with this method, but may be John, Hans or somebody else has tried this already.
When I understand you right, you do not have overlap.
Heinz, Hugh has 8 images around (45 degree yaw increment). 6 images around is usual and that gives enough overlap. Perhaps you thought Hugh meant 45 degrees tilted down. That seems unlikely. I offered some PTGui processing suggestions over on the PTGui forum, where this was cross posted.
Yes John you are right about 45° around.
Thanks for the pointers on the PTGui forum
Heinz, I moved both the lower rail and the top rail for the D300, although the top rail only moved about 1mm, which surprised me as I thought it should actually be the same.
The problem seems to be something to do with the subject not having good definition detail, which is confusing the PTGui algorithms, but thanks to John's suggestions I think I am now getting somewhere.
Suggesting that the subject is confusing PTGui is not a criticism of the software as I have watched the development of image matching software through the 1980s and 1990s and think that what PTGui is doing when it generated Control Points is close to miraculous.
Apologies for the double posting John. I have noticed that you are active on both forums, but I though Joost might make a suggestion or two.
just cross tested lenses with DX Nikon camera bodies which should have the same upper rail settings concerning L1 and L2 concerning the wikidatabase.
D300S often changes about +3mm in URS. Sometimes +1mm cross bodies. I tried to find out principals, I found some, but I learned you have to test your own combination and take databases as a guideline where to start and work around.
I would have thought same lens and camera system has same Upper Rail setting but it goes to show that physical alignment is probably safer than relying on the numbers.
A you say, "test your own combination and take databases as a guideline".
Yes Heinz, I think I have solved the problem, but now have much work to do to achieve the results I need.
I am hoping that I can revisit the site and try different arrangements for taking the photographs to verify my thoughts as to why the problem occurred and if I can define it well enough, post my findings.