Standard Camera Calibration with Checkerboard

[Anna Spracklin]

Hello,

I am trying to calibrate three standard cameras using a checkerboard, and am having some trouble getting a good calibration. The errors in each camera are about 1, but one of the cameras is in a completely different space than the other two (see attached photo of the 3D World View). When I try to track markers, the projection of the 3D point is not close to the actual point. I assume the problem is due to the calibration, but I'm not sure what to do differently.

Please let me know if you have any suggestions!

Thank you,

Anna Spracklin

[Beth Brainerd]

Hi Anna — Yes, it looks like your 3rd camera is not calibrating properly. Great work using the 3D World View (from View menu) to figure that out! Please send me your XMALab (.xma) file so I can play with the calibration.

Your checkerboards look quite small relative to image size. Are you using a lot of images to do the calibration? I think you’d need many images with the small checkerboard in many different places to calibrate the whole space with such a small checkerboard. 

— Beth

<Auto Generated Inline Image 1.png>

--
You received this message because you are subscribed to the Google Groups "XMALab" group.
To unsubscribe from this group and stop receiving emails from it, send an email to xmalab+un...@brown.edu.
<Auto Generated Inline Image 1.png>

[Benjamin Knorlein]

Hi Anna,

Can you send me your dataset (xma-file) too. I also would like to have a look what is going on.

Best,

Ben

[Anna Spracklin]

Hello,

Thanks for the quick response. Attached is the .xma file, and I am calibrating from a video, so I attached the numbers of the frames that I used to calibrate. 

I only really need a small portion of the camera view calibrated (roughly in the center of the images), so I figured a small checkerboard would be okay in that sense. I was also assuming that a smaller checkerboard would result in better resolution.

Thank you for your help!

Anna Spracklin

[Benjamin Knorlein]

Hi Anna,

Can you also upload the video-sequences somewhere, e.g. drive or dropbox?

PS: And you can send the link to me and Beth directly, so you do not have to share with someone else.

Ben

[benjamin_knorlein]

Hi Anna,

I checked your data and the resolution of the data you are trying to digitize and the data you used for calibration differ which is why the digitization does not work. You always have to use the same settings for all videos and images you acquired. Usually it should have given you an error and I added a ticket to bitbucket to make sure the error message will appear in the future, I digitized the videos from the calibration and they work fine.

But also some other remarks on the checkerboard calibration:

1) While it might be ok to use a small checkerboard, it is better to use one which covers as much from the screen as possible (or if you have a small one to not only cover the area you are working in). But of course it should not be too large as well that it cannot be fully visible in all 3 cameras.The goal of the checkerboard calibration is to determine the camera parameters as well as their position and orientation. The more data about the whole field of view of the cameras it has the better for that. You might even use some images which are not overlapping in all 3 cameras, e.g. where the checkerboard covers the whole frame, to make the calibration more robust. Also on another note, when you move the checkerboard make sure that you sample the workspace well. That means, use different orientations and positions of the checkerboard. You can think of the checkerboard as a plane in 3D space in which you measure along the 2 dimensions of the checkerboard. So the more of these planes you use the better.

2) A small checkerboard does not change the resolution, but the size of the checkers does, Therefore you can also use a bigger checkerboard with more checkers. One thing though which should be taken into account is that the checkers should be still well distinguishable and if the checkers are too small you will have trouble identifying the corners. Even worse they also might be less accurate than bigger ones in case they cannot be accurately detected. 

3) The drawing in world view looks correct to me. Camera 3 looks a little bit misleading as the view frustum of the camera does not intersect with the other 2 cameras. But the reason is that the plane (and view-frustum) are drawn with a fixed distance of the image plane to the camera center. If your camera is further away than this distance it looks misleading. Unfortunately there is no way to determine the distance based on the calibration which is why I set a value which seemed reasonable. One thing you can do to check the data though is to rotate the scene in the world view in a way that you look directly from the camera center to the image plane. In this case you will see that it still intersects with the other 2 cameras.

Hope the explanations make sense to you. Feel free to ask me more questions if not.

Best,

Ben