How to get Good Dataset for Hand-Eye Calibration

A high-quality hand-eye calibration depends heavily on the dataset you collect. To achieve accurate results, it is essential to gather good data, as the calibration quality relies entirely on the user.

We recommend collecting 10-20 pairs of robot poses and point clouds for a reliable calibration.

Robot Pose Requirements

The robot poses used when camera takes images of the calibration object should be:

  • Sufficiently varied, providing different viewing angles

  • Distributed across all joints, not only a subset

This ensures a broad diversity of perspectives. The images below illustrate the required pose variety for both eye-to-hand and eye-in-hand setups.

참고

Zivid point clouds are given in millimeters (mm); therefore, ensure that robot input poses for hand-eye calibration also use mm for translation.

../../../_images/hand-eye-poses-eye-in-hand.png
../../../_images/hand-eye-poses-eye-to-hand.png

If possible, the robot poses should also be:

  • Using the same robot configuration as in the final application (e.g., the configuration used during picking)

  • Within the same workspace region as the application (e.g., at the bin where picking will occur)

Calibration Object Visibility

The calibration object should be well exposed and fully visible in the field of view of the camera. If possible, keep the board as much as possible centered in the FOV of the camera. When capturing data, make sure to consider the camera’s imaging range.

참고

If using ArUco markers as calibration objects, not all the markers need to be fully visible in the FOV of the camera in every robot pose. However, the more markers are visible, the better the detection and calibration results will be.

The images below illustrate the calibration object as seen by the camera.

../../../_images/hand-eye-calibration-board-poses.png

Acquiring the Dataset

Right before running hand-eye calibration:

  1. Warm-up the camera

  2. Run Infield Correction

Use the same capture cycle during warmup, infield correction, and hand-eye calibration as in your application. Using our Hand-Eye GUI makes the process straightforward.

To further reduce the impact of temperature dependent performance factors, ensure Thermal Stabilization is enabled.

If you are having trouble capturing good quality point clouds, check How to get Good Quality Data on Zivid Calibration Object. Otherwise, continue to Cautions and Recommendations for Hand-Eye Calibration.