Re: HELP, cvRemap version 110 possible bug

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Re: HELP, cvRemap version 110 possible bug

Ed Elston
Martin, thanks for your reply but I just noticed it.  We have
stopped using 110 because of this because in 100 we can place the PP
on a target (distorted) and then verify that it hasn't moved in the
undistored image.  This does not hold true in 110 and is creating
many problems.  I haven't checked the center, but it isn't the PP.

I wish someone from the developers group would check into this and
answer because it just doesn't seem to be correct.  If it is, then a
good explanation would be very helpful.

Ed

--- In [hidden email], "mdale9_opencv" <martin.dale.1@...>
wrote:
>
> Ed,
>
>
> The docs say that after undistorting the principal point will be at
> the image centre.  If your bullseye wasn't at the image centre
before
> remapping then maybe this is the cause of your pixel shift?
>
> I'm having problems with image rectification, which uses some
> remapping function although there seem to be several (undistort2,
> undistortmap, remap,...). I'm tracing through the C code alongside
> Bouget's original Matlab trying to track down why the Matlab gives
> nice rectified images and the OpenCV gives some bizarre, over-
warped,
> off centre crud and it seems to be down to either undistortpoints2
or
> projectpoints2.  Maybe it isn't a bug, it might just be different
for
> a reason.
>
> Martin
>
>
> --- In [hidden email], "Ed Elston" <ed.elston@> wrote:
> >
> > I need to understand this problem I am seeing using 110.  I can
take
> > the same image using version 100 and get different results when
> > undistorting the image.  Here is a description so please comment
on
> > this.
> >
> > I target the principal point cx,cy in the distorted image on a
> > bullseye that can easily be identified.  Then I undistort the
image
> > using 100 and the bullseye is at the same precise point.
> >
> > Doing the same operation with 110 and the exact same intrinsic
and
> > distortion parameters the bullseye moves about 20 pixels.  
> >
> > Based on my understanding this cannot be correct.  Everything
points
> > to the fact that the principal point (cx,cy) should not move
when the
> > image is undistorted - CORRECT????
> >
> > Ed
> >
>