Invalid multi-sensor calibration with reduced sensor resolution

[quepas (Patryk Kiepas)]

We are using RecFusion Pro to perform a multi-sensor calibration with three Orbbec Astra Series sensors. The calibration performs well when all sensors are using the maximal resolution available for both, the depth and the color, streams:

However, when we lower the resolution of depth and/or color stream the calibration performs very poorly. It’s almost like there is none! All segments are detached by half a meter. We have tested many different settings, and none of them give sensible results:

I see reasons why lower resolution should perform worse, like it is harder to position the calibration marker in space from pixelated images. However, I don’t understand why it results in such a bad calibration? Is there a particular reason or assumption behind the calibration algorithm?

[Schwer (Alex)]

Hey Patryk,

sry for the late reply. What is going is hard to tell with the provided data. The only thing that is clear to me, is that the sensors don’t seem to be calibrated (well).
It is possible that our marker detection is not able to detect the markers anymore, if the marker is represented by a very small pixel count (due to the small resolution). Providing the calibration images could be helpful to tell a bit more. 320x240 is not a good resolution though, for similar applications we usually us the 640x480 or higher. If the marker detection works on those smaller resolution images, I don’t see other obstacles for the calibration to not work.

Best regards
Alex