On 30 Jan., 20:44, torger <[email protected]> wrote:

> However, I worked this evening with this on a few images and tested
> several methods and have great help of your input. I get good results
> now (often subpixel fits), and consistent enough.

Nice to hear you've figured out a work flow which works for you :)

> The interesting fact is that setting a 1 - 2 degrees FOV is required
> for the best results. I've tried to let Hugin optimize it by itself
> but it rarely goes down to those low values, even when starting with
> the calculated 5.42 degrees.

Hugin can only come up with a correct fov value if you do a full
spherical panorama. It certainly can't even come near to a correct
value if you only give it a few CPs. If you want to calculate an
estimate for your lens' fov, you can derive it from the distance of
your NPP to the object.

It might well be that with the specialized optics you use, treating
the incoming images as rectilinear isn't the best solution, even
though it is the obvious one. I had similar issues stitching satellite
images, and I also resorted to using artificial small fov values. But
you might try and use a different input projection.

> I've also tested auto control points, and it works kind of ok, but
> there's always some outliers so I think I continue do them manually,
> having relatively few points and well-defined places gives me a better
> overview of stitching result I think, but I guess it is a matter of
> taste.

The outliers aren't a problem. The optimizer is quite intelligent and
figures out which CPs are outliers, and they won't spoil your
registration, even if they show up on the preview as red and give you
a high average cp distance. And if you generate a generous amount of
CPs in the first place (like, with the CPG setting I have proposed
earlier), there's nothing stopping you from subsequently throwing the
worst ones of them away, using the 'clean control points' button or
doing it by a distance threshold.

Having many CPs allows you to calculate lens distortion coefficients,
which are essential. And you might surprise yourself by just how
accurate a fit you can get from a field of, say, 100 automatically
generated CPs per pair. cpfind is really very good, give it a shot.

> I normally don't pre-correct lens distortion, since Hugin indeed does
> better fitting result if it can correct barrel etc itself, but in this
> special case the optimizer easily goes haywire so having close to
> perfect rectilinear input seems to work better. I've also tried to pre-
> correct roll, but it is not really worth it, that correction works
> well.

The optimizer tends to go haywire if it doesn't have enough CPs. The
more degrees of freedom you give it, the more CPs you need to keep it
on a sane path.

Kay

-- 
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