Does this handeye calibration use RGB-D data? Or only 2D data?

[kptrifork]

Hi Jones,

Thank you for sharing the GitHub repository.

I'm quite new to the concept of Hand-Eye Calibration, even though I have a degree in Robotics Engineering.

Does your code include 3D data point clouds, or does it only work with 2D images? I was under the impression that hand-eye calibration could only be done with a 3D sensor. Am I mistaken about this?

If you have any articles or resources that could help me better understand this topic, I would greatly appreciate it.

Best regards, Kevin

[JonesCVBS]

Hi Kevin,

This code is only using 2D images and actually it´s much more common (from what I've seen) to do hand-eye calibration with 2D images. At the time I also tried using point clouds since the camera I was using also gave point cloud information, but I actually found the calibration using 2D images to be better.

Here's a good video with an explanation on the topic (You don't necessarily have to watch the second half as it is just a demo of how to use the plugin): 3 - MoveIt - Easy Hand Eye Calibration with MoveIt

Overall I used 2D images because:

1. Point clouds were much nosier than 2D images.
2. There are a lot of libraries already for 2D image calibration.
3. Calibration with 2D images requires very little setup since all you need is a chess board.
4. Calibrating with the point cloud information requires you to detect 3D features in the point cloud which can be difficult/computationally complex.
5. The 2D results were very accurate.

You can find a direct comparison in this study between 2D and 3D. It might be a bit outdated but I haven´t kept up with new studies: Hand-eye Calibration with a Depth Camera: 2D or 3D?

Best, João