Conflicts with Optimizer Info

[Ed]

Good Evening:

I'm trying to stitch an interior view and cannot figure out why the Optimizer gave a "very good" declaration (average distance less than 1.0; maximum distance not reaching 2.0) and there are still visible seams in the final output. Using a NN5 with a Canon 24mm TS-E lens on a 1Ds Mark II. This is from an interiors project that I've been doing off and on since June. I experienced trouble back then and John Houghton nicely pointed out my NN5 was off slightly, so I re-calibrated the whole rig and it's worked well since. At least using the 15mm Fisheye it has. 

Yes yes, I know I know, I've been yelled at before here for using a shift lens, but I have had success with it in the past too, especially being able to minimize ceilings, and I don't think the lens was shifted for this shot all that much. Right now I'm just confused about Optimizer information that I thought would lead to an effortless stitch and then seeing at the results that came out.

Work files and the stitched panoramic are at      http://www.edsewell.net/pano/Hadron.zip

Apologies for not being able to send the PTGui work file, but after messing with it for an hour, I threw it away and figured I'd start all over again. Then I though of all of you good folks. Any suggestions are greatly appreciated.

Ed

[John Houghton]

Ed, I didn't find any problems with the stitch, which I have uploaded
to

http://www.johnhpanos.com/Hadron2.zip

If you have a good optimizer report AND you have a good distribution
of control points, then the stitch ought to be good. I can only
assume that your stitching errors were the result of not having
control points in those areas.

John

[Michel Thoby]

Hi John,

Le 14 nov. 2012 à 08:28, John Houghton a écrit :

Ed, I didn't find any problems with the stitch, which I have uploaded
to

http://www.johnhpanos.com/Hadron2.zip

If you have a good optimizer report AND you have a good distribution
of control points, then the stitch ought to be good.  I can only
assume that your stitching errors were the result of not having
control points in those areas.

John

There are indeed some problems on Ed's panorama: all are located not far from the top and bottom of the panorama and on each and every seam between adjacent images but they are "small" and need some zooming to view them.

As you suggested, getting rid of these stitching errors is actually possible by putting much more CP very well spread over the overlap and with some of them in the flawed areas. After optimization (with three coefficients), the CP distance maximum errors then subsequently become higher (~4 to 6). While the RMS is still Very Good (~1), the few concerned CP yielding higher CP distance errors are not located where the stitch errors were located on Ed's image. The errors become invisible on the output image though.

BTW PanoTools Optimizer gives a very similar result with even higher CP distances error figures. That is expected and is not a problem.

I tend to conclude that the  EF 24 mm TS-E lens has a very small distortion (a, b and c are tiny figures) but it's an intricate radial mapping (most probably wavy). That is the case for the majority of modern wide-angle lenses with a retro-focus design (that is to say : designed for Single Lens Reflex  cameras). A rectilinear lens polynomial correction equation with a reduced number of terms has a hard time to accurately convert the lens actual projection into the ideal central perspective projection. But in practice and in this case, this is not absolutely required as the stitch errors are invisible:)

Michel Thoby

[John Houghton]

Michel, Inevitably, parallax isn't perfectly corrected either, and
that has an impact on the optimization result:

http://www.johnhpanos.com/Hadron2.gif

The parallax shifts are very small, however, and not really a concern.

John

[Ed]

John & Michel:

Very early here and I'm loading gear for another (non-pano) project. Thanks for looking at this stuff so quickly, but I'm writing right now on an iPad and won't have a chance to look at your files for a few more hours. I'm anxious to see what y'all have done (John, you have always offered help with my problems). Michel has noted what I was upset about, the seam misalignments most visible in the ceiling grids and the carpet, but for the most part, all through the final stitch. I did add numerious CP pairs all over the ceiling and that's when the Optimizer gave its so "optimistic" reading.

I'll look your files later today. Thanks, I'll be in touch and should anyone else want to chime in, please do.

Ed

[PTGui Support]

Hi Ed,

I think I've said it before but stitching will really be much easier if
you do not shift the lens. Shifting the lens also shifts the optical
center, causing a small amount of parallax. You would have to keep the
front of the lens at a fixed position and shift the camera to prevent
this. Once there is parallax you will never get a perfect stitch.

If you need straight lines for architecture scenes: this can be achieved
using any lens.

If you think the optimizer misreports the actual quality of alignment,
please post the project file here so I can take a look.

Joost

[Ed]

Joost:

Done & Done:  No more shift lens. But this is confusing because when I shift the lens for a shot like this, I only move it up or down. I never thought this would have a big effect on accurate, parallax-free capturing, since the lens isn't moving in or out, right or left, just rising or falling over the same rotation point. It has that big an influence?

I'm sorry but I can't forward the work file because even though the Optimizer said it was really good (in my limited experience, I've not seen Optimizer numbers that small too often) I got frustrated trying to fix it over and over, so I dumped it and figured I'd start from scratch. 

Ed

[Jim Watters]

On 2012-11-15 7:33 AM, Ed wrote:
> No more shift lens. But this is confusing because when I shift the lens for a
> shot like this, I only move it up or down.
>

> Ed
>
If the front of the lens was fixed and you raised and lowered the back then
there would be less of a problem. This still gives limited field of view. For
wider and/or taller panoramas it is best to rotate and tilt the camera and lens
around the NPP (No parallax Point).

--
Jim Watters
http://photocreations.ca