A tour of mrcal

mrcal is a toolkit that provides improved methods for making and using camera models (calibration, tracking, mapping, photogrammetry, etc).

The best way to convey a sense of the capabilities is to demonstrate some real-world usage scenarios. So let's go through a full calibration sequence starting with chessboard images, and eventually finishing with stereo processing and triangulation. This page is a high-level overview of all the techniques that make mrcal different and better than other tools. For more details, please see the links in the main documentation. For practical details, please see the how-to-calibrate page and the recipes page.

All images in these articles have been gathered with a Sony Alpha 7 III SLR. I want to stress-test the system, so I'm using the widest lens I can find: a Samyang 12mm F2.8 fisheye lens. This lens has a ~ 180deg field of view corner to corner, and about 160deg field of view horizontally.

In these demos I'm using only one camera, so I'm going to run a monocular calibration to compute the intrinsics (the lens parameters). mrcal is fully able to calibrate any N cameras at a time, I'm just using the one camera here.

The tour is split over a number of pages:

  1. We gather calibration images, and perform some initial calibrations
  2. Differencing: we compare several of the calibrated models
  3. Uncertainty estimation: we compute the projection uncertainties of the models
  4. Cross-validation: we compare results of two different calibrations to gauge solution quality
  5. We discuss the effect of range in differencing and uncertainty computations
  6. We use the uncertainty analysis to find the best chessboard-dancing technique
  7. Stereo processing
  8. Triangulation