On Tue, 2010-11-16 at 18:35 -0800, Radu Bogdan Rusu wrote: 
> Hi Bill,
> 
> Great work!

Ivan gets the credit for this, I just helped with the rubber cement.

> A few suggestions:
> 
>   * use a larger calibration pattern that covers more of the image
>   * use a calibration pattern with more squares

The glass we are using is slightly reflective to IR but it is
manageable. However, the transparency film we are using to make the
squares is much more reflective and we had to cut out the individual
black squares. We are probably going to look at silk screening the
pattern on to a larger sheet of glass to improve alignment.

> * if the estimates of the corners (after subpixel refinement) are still bad, an iterative fitting process (or LMedS or 
> RANSAC) could help

We will make a bag file of our calibration tests tomorrow to help move
this forward.

> Alternatively, a calibration plate with circles would be better. We already have code ready that we need to commit for 
> finding out centers of circles/ellipses and using those for calibration plates a la 
> http://www.mvtec.com/halcon/applications/imgs/3d-vision-example-3d-calibration01.png, so this might improve things.

Can this code be made available somewhere so we can test it before
resorting to silk screening?