Calibration targets at true scale

Create a calibration target at an exact size, print it directly from the browser, and verify its scale with the reference ruler.

Task

For standard calibration, start with a checkerboard: it is straightforward and provides many precise corners.

Target type

Parameters

How to choose the size. The square size sets the physical scale, while the number of squares determines the number of reference points. More points generally help calibration, but a pattern that is too fine is harder for the camera to resolve. The “fit” button selects the largest usable grid for the sheet and the available set of unique markers.

Square size

Enter this value in the calibration software

Squares horizontally

pcs

Squares vertically

pcs

Sheet

Preview

Inner corners

—

Target size

—

Status

—

File and OpenCV parameters (Python)

—

Print at 100% scale, with no margins, and with “fit to page” disabled.

In the Chrome, Edge, or Firefox print dialog, select a printer or “Save to PDF”; Acrobat is not required.
The paper size must match the sheet selected above.
After printing, measure the reference ruler at the bottom of the sheet: it must be exactly 100 mm. If it differs by even 1 mm, print again; do not compensate by recalculating the dimensions.

Why a correction factor does not work

A printer can distort scale unevenly across the sheet, so a single correction factor will not fix the target.

How it is calculated

The tool generates a calibration target of exact size and prints it straight from the browser: a checkerboard, a circle grid (symmetric and asymmetric), a ChArUco board and a ruler for scale verification. All dimensions are set in millimetres; the target parameters and ready-to-run OpenCV code are printed alongside so that the calibration software receives the same numbers.

Target types and geometry

Checkerboard: square side in mm, number of columns and rows of squares. The calibration software takes the number of inner corners — one less along each axis.
Circle grid: centre-to-centre pitch; the circle diameter is half the pitch. Number of centres per row and number of rows.
Asymmetric circle grid: adjacent rows are shifted by half a pitch; within a row the pitch is twice the row spacing.
ChArUco: a checkerboard with coded markers in the white squares; a marker occupies 70 % of the square side; the marker dictionary is chosen by square count so that enough identifiers exist.
Ruler: reference graduations to verify that the printer did not rescale the page.

Worked examples

Example 1. Field of view 100 × 75 mm. A 9 × 7 checkerboard with 10 mm squares gives 8 × 6 inner corners and occupies 90 × 70 mm — almost the whole field, while each square gets enough pixels with any camera from 1.3 MP.

Example 2. Field 500 × 400 mm and a 2448 × 2048 px camera: about 4.9 px/mm. A 25 mm square spans 122 px — robust corner detection; a 19 × 15 board with 25 mm squares covers 475 × 375 mm.

Example 3. The target is printed on A4 with “fit to page”: the ruler reads 97 mm instead of 100. Print again at 100 % scale, otherwise the calibration receives wrong dimensions.

Common mistakes

Printing with page scaling: every dimension shrinks by a few percent.
Passing the number of squares to OpenCV instead of the number of inner corners.
A target that covers a small part of the frame or is captured in one pose only: the calibration does not cover the field edges.
Glossy paper or lamination: reflections wash out corners and circle centres.
Squares too small for the chosen resolution: corner detection becomes unstable.

Questions and answers

Which target should I choose?

A checkerboard for most tasks. A circle grid when centre accuracy under perspective matters. ChArUco when the target may be partially occluded or captured at a steep angle: the markers identify every square.

How do I verify the scale after printing?

Measure the printed ruler with callipers or a steel rule: the reference distance must match the printed label. If it differs, disable scaling in the print dialog.

How large should a square be?

Large enough that it spans at least a few dozen pixels in the frame, while the whole target fills a substantial part of the field of view. The resolution on the object can be computed in the optics calculator.

What is the OpenCV code next to the target?

A ready Python snippet for OpenCV 4.13 with the parameters of this exact target in millimetres. Replace the image file name and run it.