Ctrl-Points and Distance - help me understand

103 views
Skip to first unread message

George R

unread,
Apr 18, 2012, 11:54:24 AM4/18/12
to hugin and other free panoramic software
I have just returned from a trip where airline baggage restrictions
meant I had no tripod nor pano-head with me. So I tried the hand-held-
with-plumb-line method.

I tied a weighted string around the pivotal point of my fisheye lens
and shot 8x45° instead of my usual 6x60° ... easier to guess, more
overlap and more latitude for error.

The first couple of panoramas have stitched OK ... not MUCH different
from my usual tripod mounted method.

However I am finding that some control points are being allocated a
large "Distance" in the range 60-100 even though I can SEE that the
control point in each shot, is set on exactly the same grain of grit
on the pavement or bolt in the fence, or whatever.

Usually, on my tripod mounted panos, I can get most control points to
a distance of under 2 with a few - the extreme "rogue" points - maybe
at a distance of 10 or 12. This time the highest distances are all on
points that match with the down shot, which is probably the one with
the least accurate camera positioning.

So my questions are about the nature of control-point-distance and
what I can or should do about it.

As I understand it the "distance" means something like "how far apart
those points in the images would be when each of their parent images
is projected onto the sphere that makes up the whole panorama". Is
that in any way close to what it really means?

Does the processing at the stitching stage use the distance measure to
transform the images and make the points match? In that case is a
high-distance point - that is known to be an optically-accurate-match
- useful and worth having?

Or should I just delete of high-distance points?

Does a point with a high distance, that is known to be an optically-
accurate-match, reflect camera movement away from the pivotal point?

Is there anything I can do to "fix" one of these high-distance
points. My intuition was that this was where "Translation" was going
to be helpful ... but when I tried doing Optimise with Translation I
got the effect that I always get any time I use Translation and the
preview changed to a roughly square image with all the source images
stacked on top of each other.

Anything helps me to better understand the nature of Control Points
and the use of Optimise-Translation would be very welcome.

George

Bruno Postle

unread,
Apr 18, 2012, 3:17:46 PM4/18/12
to hugin and other free panoramic software
On Wed 18-Apr-2012 at 08:54 -0700, George R wrote:
>
> Does a point with a high distance, that is known to be an
> optically-accurate-match, reflect camera movement away from the
> pivotal point?

Yes, parallax problems show up on objects near to the camera, such
as the ground. If you only pick points on objects that are a very
long way away then you get tiny distance errors for the points, but
nearby objects can still be 'chopped' when stitching.

If the distant objects are irregular then you can sometimes use just
nearby objects for control points and get a cleaner stitch with less
noticeable errors.

Otherwise you can use masks to control the seam placement for the
near objects, or fiddle with the errors afterwards in an image
editor.

Optimising d,e translation might work and usually gives a lower
error number, but I'm not convinced it delivers better pictures.

If the ground is flat, then you could use XYZ parameters to stitch
the ground separately from the rest of the scene.

--
Bruno

Reply all
Reply to author
Forward
0 new messages