[ptx] Conserving verticals with many control points

[Damien Douxchamps]

Hi Rik,

Thanks for your continued patience :-)

On Mon, 2006-04-17 at 05:14 -0700, Rik Littlefield wrote:
> Damien,
> 
> Is your "wavy horizon" a smooth sine curve with one high and one low
> spot opposite each other on the 360 degree horizon?  Or is it kind of
> wiggly, with multiple cycles and/or irregularly spaced high and low
> spots?

Yes, it's a single-period sine curve. 

> If it's a single sine curve, then it's just unlevel, in which case a
> couple of well-placed vertical controls really should take care of the
> problem.  If it's kind of wiggly, with no significant component of a
> single-cycle sine wave, then you've probably got the many-small-errors
> problem that will be harder to deal with.

It looks like you've nailed the problem :-) But unfortunately I can't
get anything from the vertical control lines. I tried a different set
and it remains the same.

> If I remember correctly, the panotools optimizer uses horizontal
> offset in the pano as the error value for vertical controls.  You
> should be able to look at the error report to see how well these are
> being pushed to zero.

The distance reported for the vertical control points after optimisation
is either 0.00 or 0.01. Only 2 decimals are shown in hugin but it looks
good already.

> >From your email, I can't hear what's going wrong.  Perhaps you could
> post out an image showing the problem?

If there's any interest I can post the whole thing (jpg + pto). Probably
in the 15 MB.

> Also, the panotools wiki has a page called "Leveling a Finished Pano"
> that talks about several options for leveling.  You might read that
> for ideas -- most of them apply even when you have multiple input
> images.  You might also try exactly what's described there, to try
> straightening your wavy horizon by remapping the finished pano.  That
> takes all of your ordinary control points completely out of the
> problem.  See 
> http://www.panotools.info/mediawiki/index.php?title=Leveling_a_Finished_Panorama

Thanks, good idea. I'll dig a bit into that because it does not work on
my first try: the wavy pano is only rigidly moved around (at least with
hugin). I've also tried to do a second optimization step with only the
pitch and roll parameters (forced to zero before starting the
optimization). It's better, but not quite straight yet.

<newbie>
There must be something obvious I'm missing...
</newbie>

Damien

-- 
  _    Damien 'Takahara' Douxchamps, PhD
 ('-   Post-doctoral investigator
 //\   Image Processing Group, NAIST
 V_/_  http://chihara.aist-nara.ac.jp/