< Previous | Contents | Manuals Home | Boris FX | Next >
SynthEyes 2024 Camera Calibration Manual
(SynthEyes 2023.10.1057)
©2017-2023 Boris FX, Inc.
Precisely matching original photography requires precise information about the camera. Most computer software is based on the simple and efficient linear perspective camera model, which characterizes the lens with a single field of view (or the equivalent focal length and required exact sensor or plate size the latter of which you rarely know! ). But real lenses produce distortion, and for an exact match-move we need to be able to understand and predict that distortion, as well as know other parameters such as the location of the lens's optic center on the image.
We'd like to tell you that you can take a real shot, and only that live shot, push a button, and have a complete set of lens parameters accurately determined from the shot alone (ie as part of the match-move). That's not the case, it's not mathematically possible: the required information doesn't necessarily exist in that shot. It might, but might not (especially for tripod shots!).
We'd like to tell you that someone can shoot a lens grid sometime, and you can later push a button and have all the necessary parameters determined. Again, that's not the case though depending on the situation you may be able to get a usable result from that.
We'd like to tell you that there is even such a thing as a list of numbers that can be figured out to describe a lens, and once that's been done, we'll know everything about it. Again, not the case. Quadratic coefficients, cubic coefficients, quartic coefficients... keep going. Inside the tube you call a lens are many simple lenses (5-15, say), of different materials. The characteristics of the overall lens are numerically simulated, and the properties of the simple lenses adjusted until the simulated result sufficiently matches the objective ("a 50 mm lens").
The actual characteristics are a complex jumble described only by that simulation. There's no simple description. Then the lens is put together with varying manufacturing accuracies to produce the lens on your camera, which hopefully resembles the simulated design.
We'd like to tell you that you can take a calibration image for a lens, then be able to use that calibration for every other combination of focal distance, iris, and zoom (if applicable).
Again, not the case. As the various elements move inside the lens, the resulting distortions do also. As a bonus complication, the apparent lens characteristics can change based on the shooting mode of the camera, as it uses different subsets of the pixels on the sensor.
We'd like to tell you that when you use lens calibration results, you'll always get nice accurate results. Again, not the case. Depending on the situation, the calibration tools can provide results that are very precise, produce low apparent errors, but in fact are giving very precise results that embed systematic errors into the lens calibration results, such that the result is less accurate in general use than if you'd done something else or simpler. This is especially true if you try to use lens calibration grids that don't extend to the edges of the image.
So from this you should understand that there are no exact answers here. The objective is to produce results that are sufficiently accurate for what you need to do to your specific shot.
Depending on what that shot is, and what you need to do, you may want to undertake various calibration activities with the live camera gear. More effort can provide better results.
And if you just get handed something, the objective is to do as well as possible, with what has been provided.
SynthEyes can calibrate based on four fundamental approaches, working from several kinds of imagery:
1) By making straight lines in 3D straight in 2D
a) from collections of indicated straight lines
b) from lines automatically generated from a grid of 2D points
2) From features with known coordinates
a) features arranged in a fixed regular grid, all exactly on the same plane.
b) features with known XYZ coordinates (not located on a single plane)
3) From a nodal shot of a dense collection of random dots or features, preferably panning more than 360 degrees.
4) From a single image of a lens grid, constructed and shot properly, even for challenging unusual lenses, such as anamorphic lenses that aren’t well-aligned.
We'll also show how to determine lens center based on (created) vignetting.
NOTE : SynthEyes can calculate cubic and quartic distortion coefficients on "live shots" that can adequately describe fisheye lenses when the image is a cropped portion of the overall fisheye image.
The calibration process for methods 1-3 above is driven by a common Lens Calibration script, accessed by Lens Master Calibration on the Shot menu. Method 4 is performed by Shot/Rectify Grid.
This manual describes each of the lens calibration techniques, including how to shoot for it. Then the Lens Master Calibration script itself is described, which not only controls the calibration process, but determines what calibration presets and maps are produced as a result. Those results depend on what workflow you desire.
The Lens Calibration script uses a different set of lens models than the newer set on the Solver panel, and is mainly intended to generate STMaps as output. However, STmaps can be cumbersome and aren’t universally loved.
To use all the new lens models and produce numeric results, you can use the Lens/Grid Calibration Setup script. It sets up a Refine solve for a lens grid; you can configure the solver with whatever lens model you want to use, with whichever coefficients that you want to be solved (or not solved). Read the section for what to do next.
Finally, there’s a section on Rectify Grid and its controls, which forces grids to be straight, producing an STmap, at the risk of burning in shooting and grid errors.
At the end of the manual, there's information on some special lens types and a Shooting Guide for Lens Calibration Grids that you can give to the responsible parties.
©2023 Boris FX, Inc. — UNOFFICIAL — Converted from original PDF.