Abstract
A method measures the lens distortion of a camera that maps image positions onto a surface, such as a fixed camera tracking performers on a stage, without a calibration target. The operator indicates a feature that is straight in reality, for example a stage edge, tape, truss or LED-wall edge, by clicking its two ends. The software places sample points between the clicks and snaps each one across the line onto the strongest image edge, using one edge polarity for the whole line so that both edges of a strip of tape are never mixed. The operator can correct individual points and switch off hidden ones. Alternatively, a fully automatic variant finds stable edge chains, keeps those that a single lens model makes straight, and proposes them ranked by information gain. From one or more lines, a plumb-line fit estimates radial coefficients. The fit is restricted to models that do not fold inside the frame, and higher-order terms are used only when the lines determine them. The predicted uncertainty of the result is turned into a coverage rating with application-specific limits, together with a hint naming the image region where another line would help most. The lens model is applied to the drawn overlay and to input coordinates rather than to the video, and the reference points of the plane calibration are undistorted before solving, so that the calibration can be re-solved after a lens change. Variants cover other ways of seeding and locating lines, other lens models, quality measures, outputs and implementations.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Honold, Michel and Hermann, Paul, "Measuring the Lens Distortion of a Fixed Tracking Camera from Operator-Traced Straight Lines, with Edge Snapping, an Uncertainty-Based Coverage Rating and Grid-Side Correction", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12051