On 8 April 2011 09:38, Yclept Nemo <[email protected]> wrote:
>
> Yes I did manage to borrow a panoramic head; even without out, since
> the lightpost is about 50-75 feet distance I doubt parallax would
> cause any problems.

It should be straight-forward to calculate the maximum parallax error
(in pixels) based on an estimated upper bound in the change of
position of the no-perspective point and the angular distance between
pixels at 28mm.  But yes: I would have thought that that would be
sufficient, even if the control points were quite a lot further away
and the pano head wasn't exactly calibrated (i.e., not rotating
precisely around the NPP).

> I use a cheap Canon EFS 18-55mm lens which I calibrated specifically for 28mm

I think I would be worried about not being able to re-set the lens to
exactly the focal length it was calibrated at.  On my (relatively
cheap) lenses, even if I were to tape the zoom ring to the body so it
couldn't rotate between calibration and shooting (or equivalently make
sure it is hard against the 18mm stop), there is enough play in the
mechanism that I would not expect to get good results without
optimizing (unlinked) v.  I don't generally find it necessary to
unlink a, b, and c, though, although I do include these in the
optimisation.

Aside: I admit that I'm not really sure I see the point of being able
to save/load lens parameters when I can just include these in the
optimisation.  Is there any advantage, other than perhaps saving some
CPU time?  I would think think that, except for prime lenses, any
reduction in optimisation time would be at the risk of the lens not
actually being at the same focal length that was used for the
calibration.

> Anyway, is it possible to produce a perfectly aligned 360 degree
> panorama? I've been trying really hard - many different strategies -
> and am unable to get rid of artifacts.

I have shot and aligned only one full 360-degree panorama, and it was
from handheld, and despite this I did not have much trouble getting
things to align sufficiently for there to be no visibile artifacts.
Here is the strategy I used, which I've used before on several other
less-than-360 panoramas with good success:

- Shoot three bracketed exposures of each of 21 stacks (9 stacks
around the equator) (from the middle of a large, cobbled public
square).

- Load 63 images into Hugin.

- In the image list and in the preview, select the middle-exposed
image from each stack (#1, 4, 7, ... 61) and use cpfind to create
control points.

- Optimize (y, p, r) to start with, with "Only use control points
between image selected in preview window" ticked.

- Manual control point editing: fine-tune all; delete all control
points in sky and most on the cobblestones in the foreground (large
parallax errors due to hand-holding); delete a few more down an alley
(again, parallax errors); add quite a few manually between overlapping
images where cpfind did not do a good job (probably about 25cps on
each of a dozen different overlap pairs).

- Optimize (y, p, r, v (unlinked), a, b, c (linked), d, e (unlinked))
and continue to examine and adjust or delete poorly-aligned control
points until errors are relatively small (average error = 0.85,
maximum error < 3.8).

At this point almost all the control points are on the fronts of
buildings around the square.

- Do test stitch to check alignment.  No problems - even with cobbles
in foreground enblend has done a good job hiding misalignments.

- Now, select first stack (images 0, 1 & 2) and use Align_image_stack
linear to create control points.  Repeat for each additional image
stack.

- Create control points manually for images where Align_image_stack
did not do a good job (usually due to my having allowed the camera to
roll slightly between exposure, which usually results in control
points being clustered around centre of image with few at edges).

- Select all images in preview window, and optimize the same
parameters as before (y, p, r, v (unlinked), a, b, c (linked), d, e
(unlinked)) to bring everything into alignment.

- Stitch.  Add masks to deal with moving people, birds, etc.  Repeat.

There are probably several ways I could improve this process (e.g.:
I'm not sure it's necessary or particularly useful to select images in
groups of three when using Align_image_stack).

In general, I find the strategy of having only the middle-range
exposure from each stack linked to adjacent images, with lots of
auto-generated control points between the images in each stack
provides good results while keeping the total number of control points
to a reasonable value (~8k, for the panorama described above).


Christopher

-- 
You received this message because you are subscribed to the Google Groups 
"Hugin and other free panoramic software" group.
A list of frequently asked questions is available at: 
http://wiki.panotools.org/Hugin_FAQ
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to 
[email protected]
For more options, visit this group at http://groups.google.com/group/hugin-ptx

Reply via email to