IRvision Camera/Pickup Issues

    Hello All,

    I am trying to set up a simple 2D vision process with a fixed camera. I've trained the TCP and User Frame correctly: TCP is just a pointer with a Z offset, User Frame has been calibrated and recalibrated using 3 and 4 point methods and both give the same values. Question: Does it matter what frame you are in while training the User Frame?

    My issue is when setting up my camera; I use a fixed (bolted in place) calibration grid that was also used to teach the User Frame but when I get to the calibration step if I use Perspective with an automatic camera distance calculation I get a value of around 1 m, while my measured height is 416 mm. This method gives good mean and max errors, 0.25 and 0.693 pix respectively. When I use Perspective and override my standoff distance to my measure value my mean and max errors are around 1 and 3 pix respectively. Using Orthogonal gives mean and max errors equal to those I get with my standoff distance overridden.

    My camera is a Basler acA1920-48gm with a Fujinon 1:1.6/35mm lens.

    All three calibration methods (perspective auto, perspective override, and orthogonal), when used in the same vision program give about a 20 mm offset from the part.

    Had trouble uploading pictures but they can be found at these links:

    20230608-121254 hosted at ImgBB
    Image 20230608-121254 hosted in ImgBB
    ibb.co
    20230608-121313 hosted at ImgBB
    Image 20230608-121313 hosted in ImgBB
    ibb.co
    fanuc-error2 hosted at ImgBB
    Image fanuc-error2 hosted in ImgBB
    ibb.co
    Fanuc-error1 hosted at ImgBB
    Image Fanuc-error1 hosted in ImgBB
    ibb.co

  • 95devils June 8, 2023 at 9:28 PM

    Approved the thread.

    Question: Does it matter what frame you are in while training the User Frame?

    There are user , tool , and jog frames. It only matters which Tool frame is active and that it is correct. If you teach a user frame "grid frame" with incorrect tcp then the calibration will never be accurate.

    One standoff is 416mm the other is 1191. Something not right here.

    Maybe not a true solution but I solved the error by creating a new 2D camera and 2D vision process. I still had the discrepancy between automatic camera calibration and override standoff distance ( 1119 mm vs 416 mm). Orthogonal gave similar results to my 416 mm standoff so I decided to go with Perspective with the standoff distance overridden to 416 mm.

    This did give me a mean error of around 1 pix and max of 3 pix but since I am using 2.3 MP camera on a FOV of around 100 mm square, a max calculated error of 0.15 mm is ok.

    I also found that changing the center origin point in the vision process removed the slight offset I was seeing on pickup.

    Hi,
    As mentioned, when teaching a User Frame, all that matters is that the Tool Frame currently active is accurate.

    Next, Orthagonal calibration is only used when the optical axis of the camera is dead straight perpendicular to the calibration plane. This can be achieved by placing a (flat) mirror underneath the camera, activating the crosshair in the view, turn on Live view and then adjusting the position of the camera until the reflection of the camera lens ends up perfectly in the middle of the picture.

    Otherwise, use perspective.

    Choose an Application frame that has the Z-axis parallell to the camera's optical axis. (If the camera is pointing straight down and the robot is mounted with 0° mount angle, then use UF[0] World as Application Frame).

    Next, make sure to use the 4 point method if using Calibration Grid method. Put Orient Origin in the bottom right dot, put X-point in the top right dot and Y-point in the bottom left dot. Then put the fourth (System Origin) point in the middle point of the grid. That way you will get a User Frame with the same orientation as the grid and with it's origin in the middle of the grid.

    Make sure to specify in the calibration process the correct distance between the dots on the calibration grid! If not, the the conversion between pixels and mm will be off!

    Regarding your standoff values, try overriding the focal length and input the value of your lens (35 mm?).

    Also, when using a vision process with Fixed Frame Offset, specify a User Frame with origin close to the center of the Field Of View to get better accuracy (suggestion is to use the same User Frame as you used for the calibration grid).

Advertising from our partners