Influence of image selection on calibration results

[bijann...@gmail.com]

Hello,
I use the tool box to calibrate fisheye cameras (Mobotix Q25 6MP). For this I use a testand, were I Can adjust accurately the relative position between the camera and the checkerboard pattern. Images at the same angular position can be reproduced via markers at the teststand. Calibrations from different data sets with the same image positions lead to comparable results. However, if the weighting of the number of images in an area of the camera image is changed, there are significant differences. When I compared two calibration sets from the same camera I observed angular differences up to 2.5° (at zenith around 0° increasing towards the horizon). These deviation result from a elevation angle mask created with the ocam_model. The only difference between both data sets is, that one set has more images close to the horizon. Both data sets include valid checkerboard images for all sections of the camera image. Do I have to consider any other special rules when selecting images? Are the deviations to be accepted as typical uncertainties? If helpful, i can share my data set and a description of the teststand.
Thank you and best Regards