Hi,

in general, the more calibration parameters you use, the less stable your estimated parameter values will be. Some parameters like to compensate each other (e.g. distance from the scene with focal length for nearly planar scenes). The fact that your parameter values differ a lot does not necessarily mean that your calibration failed, it signalizes that your calibration pattern and/or acquisition setup cannot reliably estimate the true values of the parameters you requested (but the reprojection error is still small for all the measurements you have). For example, if you do not have a sufficient number of measurements near the borders of the image, radial distortion may be negligible and therefore hard to estimate.

All this does not need to be a problem if the acquisition setup and the type of captured data (depth variability, coverage of viewfield, ...) you take during the operation mode will be close to what you used during calibration. In that case, all (any of) the estimated parameter values should model a similar projection in the respective work range. If the calibration setup is significantly different (e.g. not as diverse as the operation mode is), you can easily run into trouble as different estimated parameter values will extrapolate the projection very differently outside the original (calibration) work range.

Best regards,  Michal.

Dear  Saulo,

Try to limit your calibration model to the following parameters:

xp, yp, c, and K1 (first parameter of the radial distortion) and check your results. Then upgrade your calibration model gradually.

Regards,

Gamal

Thanks for the replies, fellows.

Sampath: Regarding the pattern, I observed the "unstable" behavior both with planar and 3D patterns.

Michal: Very informative reply. What are your thoughts on the number of frames used as input for the calibration?

 Gamal: Is "c" the camera focus ?