Pano Head Alignment Question

[AndrewKulin]

Is what I am going to describe below a valid method for determining the "nodal point" of a lens with a Pano-Head? (in this case Nodal Ninja 5 and PTGUI).  I am aware of other methods (such as masking tape on window compared against a distant object) but right now the only time I can test this out is evenings and it gets dark early so I am trying this indoors

I am Shooting a static indoor scene (home office), with desk, PC monitor, chair etc. all at fairly close distance to camera/tripod (5 - 10 feet) - I am figuring this is going to be somewhat tough on basis of parallax potential

-shoot series of 4 shots (portrait orientation), varying the rail position on the horizontal arm (eg., 65 mm, then 75 mm, ..., to 135 mm)
- run the four shots through PTGUI then run its optimizer to identify average, min and max control point distances
identify point where smallest such variation and range above are measured by optimizer and call it my "nodal distance"
- I could fine tune further if necessary (e.g., for this particular lens, 125 mm gave best result based on optimizer only - I could try 120 mm and 130 mm to see if better result could be achieved, etc.).

Here are results for example (mm distance on Ninja arm followed by PTGUI's Average, Minimum and Maximum control point distance in pixels)

105 mm - 0.791, 0.129, 1.872
115 mm - 0.585, 0.043, 1.266
125 mm - 0.494, 0.019, 1.000 <-- best of all
135 mm - 0.593, 0.032, 1.229

Andrew

[OldRoy]

Is what I am going to describe below a valid method for determining the "nodal point" of a lens with a Pano-Head? ...
Andrew

My, that's a bit complicated.
http://www.johnhpanos.com/epcalib.htm
This page offers some useful suggestions and links by someone very competent in this area. Personally I've found the "gunsight" method easy and effective (I use a ballpoint pen tube as a sight).

Getting the NPP set accurately (within the inherent variation possible using Fisheye lenses) only seems to be really critical - IMHO - when trying to do 360 x 180 VRs with a minimum of intervention at the stitching stage - I too use PTGuiPro.

Apologies if I'm misunderstanding your question.

Roy

[fike]

That seems like a difficult way to do it, but it would work, I guess.  What lens are you using?  Wide angle or normal?  You should be able to setup a small range that would work. I have used a ruler placed horizontally on a shelf with something vertical crossing it in the foreground.  If you are using a wide angle lens, you shouldn't need a huge distance between foreground object and background object.  I have done the work with the camera 5 feet from foreground object and then 5 more feet to background object.  If you have a decent LCD on your camera, you should be able to do it all with live preview zoomed-in.

[Panopeeper]

This is not only an overly complicated and tiresome way, but it is not reliable, because parallax error is only one source of control point mismatch. Lens geometrical distortion and the incorrect specification of the HFoV cause mismatch; the latter is particularly a problem.

Simply comparing the distance between suitable objects in the two shots (one with the camera turned to the left, one to the right) should suffice. Note: eyeing through an optical viewfinder is acceptable for the first adjustments, but not for the final judgement, because the eye's position over the viewfinder changes how you see the image.

[AndrewKulin]

My, that's a bit complicated.
http://www.johnhpanos.com/epcalib.htm
This page offers some useful suggestions and links by someone very competent in this area. Personally I've found the "gunsight" method easy and effective (I use a ballpoint pen tube as a sight).

Getting the NPP set accurately (within the inherent variation possible using Fisheye lenses) only seems to be really critical - IMHO - when trying to do 360 x 180 VRs with a minimum of intervention at the stitching stage - I too use PTGuiPro.

Apologies if I'm misunderstanding your question.

Roy

Roy I am aware of the site and have used the tape on the window technique in the past with other lenses.  My situation is I am off to Algonquin this weekend and have picked up a new lens and want to get the pano set-up sorted before traveling up there, while juggling work, family and volunteering commitments - so first chance I had to do anything was last night about 11 PM and tried what I described which I thought had the advantage of being in a lighted location, and being next to my PC allowed me to quickly load the shots into PTGUI run the stitch and view the end result and optimized results quickly.  So I thought of this as a shortcut to some of the other techniques available.

I suppose the crux of my question is, if in running PTGUI and looking for the distance on the slide rail (currently about 125 mm) where the difference between the auto-generated control points reaches a minimum, is that effectively my lens' nodal point?  I think it is but looking for confirmation from those with more experience in panos.  

For what it is worth, at 125 mm, the red ring on the lens I am testing in this manner looks (by eye) to be centred over the pivot point of the pano-head.

That seems like a difficult way to do it, but it would work, I guess.  What lens are you using?  Wide angle or normal?  You should be able to setup a small range that would work. I have used a ruler placed horizontally on a shelf with something vertical crossing it in the foreground.  If you are using a wide angle lens, you shouldn't need a huge distance between foreground object and background object.  I have done the work with the camera 5 feet from foreground object and then 5 more feet to background object.  If you have a decent LCD on your camera, you should be able to do it all with live preview zoomed-in.

The lenses I use , or plan to use, have been my 17-55/2.8 (at either 17 or 55 mm), my 70-200/2.8 (at 70 mm), and my new 24 mm TS-E/II (on a 40D)

This is not only an overly complicated and tiresome way, but it is not reliable, because parallax error is only one source of control point mismatch. Lens geometrical distortion and the incorrect specification of the HFoV cause mismatch; the latter is particularly a problem.

But in terms of identifying the nodal point for my lens using this technique for future reference, this technique or any of the others identified in the link posted by Roy should be sufficient, correct?

That is interesting about distortion and the HFoV.  What do I have to look for in regard to HFoV and how do I get a correct specification for it?  I presume by your answer that simply letting PTGUI work this out automatically is not good enough?  Is there a link or reference that you can provide on the subject?

Andrew

[Panopeeper]

But in terms of identifying the nodal point for my lens using this technique for future reference, this technique or any of the others identified in the link posted by Roy should be sufficient, correct?

I would not do it this way. It is very simple to line up two objects, make two shots (camera at the extreme left and extreme right position) and look at the distances between the objects. Pixel peeping. (I am doing this on the non-demosaiced raw images, so that I see even a single pixel difference.)

That is interesting about distortion and the HFoV.  What do I have to look for in regard to HFoV and how do I get a correct specification for it?  I presume by your answer that simply letting PTGUI work this out automatically is not good enough?

This is a complex issue. I don't know how far you are acquainted with the optimizing step. You should do that manually to understand it.

The optimization is controlled by many parameters:

- control points, including lines, verticals, horizontals
- angle of view (changes with the focusing distance!)
- lens distortion

The optimization process tries to locate the control points relatively to each other of the matching pairs using all those parameters, in a way that the root mean square of the control point distances becomes the smallest possible (you see the thousands of iteration steps, right?).

1. I start out with constant angle of view, and the lens distortion parameters zeroed out and fixed (I usually let the first parameter "open"). I correct the control points as much as possible (I am setting them manually).

The optimization calculates the best possible matching for the given parameters, but that is usually not good enough.

2. I "open up" the lens distortion parameter. This allows the optimizer to make the avrg control point distance even smaller by assuming lens distortion.

3. If the result is still not as good as I like it, *now* I open up the angle of view parameter. This allows the optimizer to assume different angle of view for the individual frames. This may be incorrect (if you shot all frames with the same focusing), but it does not matter if the result gets better. I shoot the frames almost always with variabe focusing, i.e. the focusing distance is not constant, thus the angle of view is not constant either.

This process leads to a warping of the source frames, which brings the control points much closer than with fixed parameters. However, the calculated values of the parameters, like angle of view, may not be correct (but who cares, if the result is good?). Thus, you can not go backwards and accept the field of view calculated by the optimizer as something "true". That may be useful (not unconditionally true) in one pano, but you may get other values in the next pano.

Facit: do not use the optimizer when determining the location of the entrance pupil. That is a relatively simple issue and does not depend on stitching technique.

Final notes:

- many (perhaps most) lenses move the location of the entrance pupil by focusing. Therefor it is not a good idea to place those objects very close, as it is in some guides,

- the EP location can be determined pretty good for wide angle lenses (I mean with one millimeter accuracy), and medium good for the middle range. However, it is not so clear with long lenses, like those over 100mm focal length. Do not be surprized when you find, that five millimeter more or less in the adjustment does not make any discernible difference. This is not a problem, for you are not shooting at close distance with a long lens, except in very special projects.