search

ethz-asl / kalibr

The Kalibr visual-inertial calibration toolbox
Other
4.21k stars 1.38k forks source link

Incorrect checkerboard pattern detection #608

Open shrutarv opened 1 year ago

shrutarv commented 1 year ago

I have used multiple combinations of checkerboard patterns for calibration. But, kalibr does not detect the corners in an ordered way. please have a look at the attached image. Is this just a visualization error? Please help me resolve this. issue Screenshot from 2023-04-14 10-37-28

goldbattle commented 1 year ago

No, this is the problem of using checkerboards (it isn't clear the ordering so you can have it "flip" mid way through a dataset). This is also why it is not possible to have a square checkerboard. See the top of the wiki page: https://github.com/ethz-asl/kalibr/wiki/calibration-targets

shrutarv commented 1 year ago

I tried with april grid. But, kalibr does not detect the grid patterns. Please have a look at the screenshot detection_issue

goldbattle commented 1 year ago

Is this a thermal camera? The contrast seems very low. Likely it is a resolution or focus problem with your sensor.

shrutarv commented 1 year ago

It is an event camera. I adjusted the focus. The image is slightly better. But, still detections are low. issue The second image is a RGB camera.

RGB camera

shrutarv commented 1 year ago

No, this is the problem of using checkerboards (it isn't clear the ordering so you can have it "flip" mid way through a dataset). This is also why it is not possible to have a square checkerboard

I also tried with a 6x9 checkerboard pattern. Please have a look at the second image. It looks like there is a bug in kalibr. Could you please hep me out here.

goldbattle commented 1 year ago

The detection of the checkerboards look ok in your original posting. What is the problem or could you clarify?

See the example here: https://docs.opencv.org/3.4/d4/d94/tutorial_camera_calibration.html image

Maybe you are referring to the colors not being per-row in your picture? This maybe is a bug in visualization or how the extractions are tracked. Is this what you mean / could you clarify? I originally thought you were talking about that detection can be on the horizontal rows or the vertical columns.

shrutarv commented 1 year ago

Yes, I meant the the colors not being per-row in the visualization. Should I ignore that and trust the calibration?

goldbattle commented 1 year ago

Very good point, should be looked into.