Cautions And Recommendations

Common mistakes

Image quality

Ensure the calibration object is well exposed and in focus in all images.
Keep the object within the depth of field:
- Use higher \(f\)-number (smaller aperture).
- Compensate with longer exposure times.
Use the same aperture setting for all images to avoid fluctuations caused by varying aperture sizes.
Capture calibration object from diverse views by exciting all six degrees of freedom of both the calibration object and the robot.

Zivid API

Zivid point clouds are given in millimeters (mm).
Therefore, ensure that robot input poses for Hand-Eye calibration also use mm for translation.
As a result, the Hand-Eye output translation values will be in millimeters (mm).

Detectable ArUco marker

If using a Zivid calibration board, ensure the ArUco marker is visible in all poses; otherwise, checkerboard detection will fail.
If using one or more ArUco markers as calibration object(s) not all markers need to be visible in every pose.

탐지 API는 탐지 실패 이유에 대한 정보를 제공합니다. 다음은 탐지 상태를 확인하는 방법의 예입니다.

C++

source

const auto detectionResult = Zivid::Calibration::detectCalibrationBoard(frame);

if(detectionResult.valid())
{
    std::cout << "Calibration board detected " << std::endl;
    handEyeInput.emplace_back(robotPose, detectionResult);
    currentPoseId++;
}
else
{
    std::cout << "Failed to detect calibration board. " << detectionResult.statusDescription() << std::endl;
}

C#

source

using (var frame = camera.Capture2D3D(settings))
{
    var detectionResult = Detector.DetectCalibrationBoard(frame);

    if (detectionResult.Valid())
    {
        Console.WriteLine("Calibration board detected");
        handEyeInput.Add(new HandEyeInput(robotPose, detectionResult));
        ++currentPoseId;
    }
    else
    {
        Console.WriteLine("Failed to detect calibration board, ensure that the entire board is in the view of the camera");
    }
}

Python

source

detection_result = zivid.calibration.detect_calibration_board(frame)

if detection_result.valid():
    print("Calibration board detected")
    hand_eye_input.append(zivid.calibration.HandEyeInput(robot_pose, detection_result))
    current_pose_id += 1
else:
    print(f"Failed to detect calibration board. {detection_result.status_description()}")

Eye-to-hand

The calibration object is mounted on the robot end-effector and moves with the robot.
Mounting can be fixed directly to the flange or held by a gripper.
Exact mounting location is not critical (it is not necessary to know its pose relative to the end-effector).

Recommendations:

Ensure the calibration object does not move relative to the flange or gripper during robot motion.
Use rigid materials for brackets and calibration object (e.g., use Zivid On-Arm Mount to mount the Zivid calibration board to the robot).
Ensure smooth robot motion with controlled accelerations to avoid shaking or dislocation.
Ensure the calibration object is still during image acquisition (e.g., wait a few seconds after robot motion to allow vibrations to settle before capturing).

Eye-in-hand

The calibration object is stationary in the robot’s workspace.
The object’s exact location is not critical (it is not necessary to know its pose relative to the robot’s base).
The camera (mounted on the end-effector) must be able to see the calibration object from multiple perspectives.

Recommendations:

Ensure the calibration object is securely fixed and remains stationary during calibration (e.g., use Zivid Stationary Mount to mount the Zivid calibration board to a workbench).
Ensure the camera is securely fixed during calibration (e.g., using Zivid On-Arm Mount).

Environment conditions

Keep the camera temperature stable during calibration:

Ensure Thermal Stabilization is enabled to minimize temperature-related performance variations.
Perform a Warm-up procedure before starting.
Keep calibration conditions similar to the expected operating conditions.

Accuracy and recalibration

Picking accuracy depends on:

Camera calibration
Hand-eye calibration
Machine vision software
Robot positioning

Robots are repeatable but not inherently accurate due to the factors such as temperature, joint friction, payload, manufacturing tolerances.

Recommendations:

Perform mastering (zeroing) to teach the robot the exact zero position of each joint.
Calibrate the robot for improved pose accuracy.
Repeat hand-eye calibration if:
- robot loses calibration.
- camera is dismounted and remounted.
Perform regular recalibration (robot and/or hand-eye) to maintain system performance.

Choosing the correct method

Avoid methods that compute or require Camera Intrinsics during hand-eye calibration:
- Zivid cameras are already calibrated with a complex proprietary model.
- Zivid SDK provides approximations for OpenCV and Halcon models, but these lose accuracy (unlike the proprietary model).

Recommendations:

Use Zivid Hand-Eye Calibration, which leverages 3D point cloud data for extrinsic calculation.

Alternative methods based on CAD matching are also ok (they also calculate camera extrinsics from the point cloud); however, they may lack Zivid-specific optimizations.

Additional advantages of Zivid method is support for ArUco markers, which can be permanently installed in the robot cell (on-arm cameras) or on the robot (stationary cameras).

계속 읽기 Hand-Eye Calibration Residuals.

Version History

SDK	Changes
2.15.0	탐지에 실패한 이유에 대한 정보가 detection API에 추가되었습니다.