Re: intrinsic / extrinsic camera calibration

Well, I'm still at it.  Today I didn't pay as much attention to the 
morning talks as I should have and did a comparison between 0.9.7 beta 5 
(cvCalibrateCamera2) and 0.9.7 beta 4 (cvCalibrateCamera) and the 
associated functions.  I did this on my own data, and on Zhang's data 
(from http://research.microsoft.com/~zhang/calib/) for verification.  And 
my results are that both beta 5 and beta 4 computed values extremely close 
(~1%) to Zhang's results.  However, for my set of input images, they 
returned very different sets of values, which I guess means that it really 
is much more sensitive than I expected to the quality of the input 
datasets.  One thing definitely worth mentioning is that Zhang's dataset 
uses only 5 images, and any subset of 4 of them will produce good, close 
values, so it would seem the acquisition of 20 or so images is unecessary. 
A better approach that I think I'll take is to acquire 5 images, and then 
compare the calibration results of all 5 vs. each subset of 4, and see how 
big the differences are.  If the results vary too much, I won't accept the 
results and instead require acquisition of better inupt images.

If anyone's interested in the specifics of how I did my comparisons, I'm 
happy to talk about them in more detail.

-stephen diverdi
-sdiverdi@xxxxxxxxxxx

>
> No replies to this?  I've noticed that this question has been asked a 
> few times in recent months, without much discussion.  At this point I'm 
> more interested in hearing if anyone has successfully used 
> cvCalibrateCamera2 for their camera calibration needs, more than 
> examining my particular case.  Are there people out there currently 
> using the new cvCalibrateCamera2 routine with satisfaction?
>
> My next step is to try doing a comparison between the results from the 
> older cvCalibrateCamera and cvCalibrateCamera2, both with my data and 
> with Zhang's example data for comparison with his results. This may take 
> me a little while, but I'll post my results when I get them.
>
> Can anyone say anything more about the algorithm used in 
> cvCalibrateCamera? cvCalibrateCamera2 is clearly marked as being a port 
> of the Matlab call based on Zhang's work, but no such attribution exists 
> in the older source.
>
> Thanks,
>
> -stephen diverdi
> -sdiverdi@xxxxxxxxxxx
>




Change settings: http://www.yahoogroups.com/mygroups, select
   Get Emails (get all posts)
   Daily Digest (one summary email per day)
   Read on the web (read posts on the web only)Or Unsubscribe by mailing 
OpenCV-unsubscribe@xxxxxxxxxxxxxxx