Camera lens distortion and calibration

Most camera lenses introduce visual distortions into the image as they’re recording and should be corrected when you are trying to make accurate measurements using cameras. OpenCV’s page on Camera Calibration has an excellent description of what kinds of distortion are found in pinhole cameras and how to correct for it. MAPLE corrects for radial and tangential distortions using these methods when the user provides dist_coeffs in their config.yml file.

Note

If you have a camera & lens setup that has relatively low distortion, you can save CPU cycles by not providing dist_coeffs in the configuration file and this will skip the undistort step when detecting Apriltags. You can see up to >2x speed improvement, which is very helpful for high FPS cameras.

How to calibrate your cameras in MAPLE

Note

If you have estimates for your fx, fy, cx, or cy parameters, populate these values in your config.yml. MAPLE will use these for its starting point when calculating the distortion coefficients. Leave these parameters blank otherwise.

  1. Print out the 9x6 checkerboard calibration target onto a piece of paper and tape it down completely flat onto a clipboard. The exact size of the checkerboard doesn’t matter, what’s important is that the paper is completely flat.

  2. Edit your config.yml file and add calibrate: true to your first camera you want to calibrate.

  3. Restart MAPLE.

  4. Refresh the MAPLE webUI. Your camera viewer should now say [CALIBRATING 0/20] in the top left corner.

  5. Hold the calibration target close to the camera so that it takes up as much of the frame as possible. MAPLE will start taking still frames every 0.5 seconds and will overlay detections when it correctly identifies the checkerboard pattern. Slowly move and rotate the checkerboard in view of the camera until all 20 calibration images are acquired.

    _images/calibration.gif

  6. Check the logfile to find your values for fx, fy, cx, cy, and dist_coeffs. Copy and paste these into your config.yml, then remove the calibrate: true directive.

    [2024-12-20_14:54:46.983] INFO:
########################################
CAM_REAR results
     Copy and paste the following into your config.yaml file under CAM_REAR to apply the calibration:

     fx: 523.534
     fy: 519.153
     cx: 320.269
     cy: 254.851
     dist_coeffs: [0.0272053, -0.213269, 0.00484789, 0.00450489, 0.16578]
########################################

  1. Repeat steps 2-6 for each camera on your robot.