MetrologyLab
.com
www.
The basic goal is to increase the accuracy of computer vision by meticulously modeling the geometry of a camera with lens, particularly with a wide-angle lens, for:
-
3-D triangulation accuracy in stereo and motion paralax calculations
-
Improving the alignment of lens elements during their assembly
-
Making more accurate measurements in photos/videos for various applications
This video shows the modeling of the radial distortion of a wide-angle camera (160 degree FOV). Each of the two final curves shown, extending into unknown territory beyond the circular AOI modeled, are modeled by a final parabola that blends into a line. In the pixel-to-mm case, the final line is essentially vertical. In the mm-to-pixel case, the slope is -4.44 because one bucket moves outward by 4.44 pixels.
Radial Distortion Model
(Click on graph below for paper)
Polynomial vs Spline Model for Camera Lens Radial Distortion
(Click on graph below for paper)
Below are some videos, presentations and notes, older first.
Cars need 3 forward-looking cameras
Saving lives via collision avoidance is a motivation for improving 3-D vision. Click on the image to read the short article via LinkedIn.
Calibrating a Tesla Model 3’s Cameras
Early work that lead to the calibration booth idea. Download this presentation with embedded videos to open it in PowerPoint. [59MB]
Camera Calibration Booth video
A future way to calibrate all of a vehicle’s cameras with 360 degree visual coverage. Click on image to view video. [410MB]
Error Vectors
Illustrates nonlinear optical errors, explaining how error vectors are computed. Click on image to view the PowerPoint presentation. [8MB]
Fisheye Lens Calibration
The presentation shows calibration of an Eufy smart door lock’s camera and a Nikon fisheye lens (actually, a Nikon Fisheye Converter that screws onto the end of the Nikon Coolpix camera, like a lens filter). Download this presentation with embedded videos to open it in PowerPoint. [107MB]
Geometric Accuracy of an iPhone 16 Pro Max Camera
521% accuracy improvement of the iPhone 16 Pro Max’s “Ultra-Wide Camera” geometry over that of the iPhone 15 Pro Max measured! Click on image to view the PDF file. [8MB]
Fisheye calibration
via images of two target types
The idea here is to compute the center of the optical axis in the top image first, resulting in a radial distortion model covering only ~2/3rds of the full FOV, the inner part. Then, use an image of a corner such as would be encountered in the calibration booth described earlier to expand the reach of the radial distortion model using the pre-computed optical center. The accuracy of results does not depend on the corner's three walls being perpendicular to each other, making the setup easy. However, the lens must have a small minimum focus distance like the Nikon Fisheye Converter used earlier or the target has to be big.
[Actually, the optical center can be calculated from the second image alone. This is how the future calibration booth will work.]
Audio/Video of the Metrology Tools
The following four videos demonstrate the three basic metrology tools: edge locator, line fitter, and arc fitter. I struggled to narrate as I moved the tools about when I created the videos in January 2024. They make up a playlist on YouTube which has better viewing controls: https://lnkd.in/gRTtCHUB
Video #1: The setup. (Duration: 2.0 min.)
Video #2: Edge locators illustrated along with their use by the line and arc fitters. (Duration: 8.4 min.)
Video #3: Line fitter. (Duration: 6.0 min.)
Video #4: Arc fitter. (Duration: 3.5 min.)
