Start with the geometric result you need
A calibration target supplies known points; the software model determines what is estimated from them. Pixel-to-length scaling in one plane, perspective correction, intrinsic camera calibration and distortion correction are different tasks. Specify the task before selecting a pattern or collecting images.
Compare detected centres and corners
Dot arrays and checkerboards can both support geometric calibration. Their practical difference is the feature detector, point ordering and response to lighting, tilt and image scale. Neither pattern is universally more accurate; the target and algorithm must work together.
| Pattern | Coordinate feature | Check before use |
|---|---|---|
| Checkerboard | Internal corner intersections | Inner-corner count, square pitch and border |
| Symmetric dot array | Circle centres in regular rows | Row/column order and orientation ambiguity |
| Asymmetric dot array | Centres in staggered rows | Exact coordinate convention, row offset and pitch |
| Resolution target | Bars, edges or radial detail | Use for detail/contrast testing, not as an interchangeable calibration grid |
Dot centres
Coordinate spacing describes centre-to-centre pitch; circle size determines how features are detected.
Internal corners
A 6 × 4 square board contains 5 × 3 internal corners. Enter the count required by the software.
Count the geometry the software expects
For checkerboards, square count and internal-corner count differ by one along each axis. For dot arrays, the diameter is not the centre pitch. In a staggered array, nearest-neighbour distance can differ from the coordinate step used by the calibration library. Supply explicit coordinates when a pitch label is ambiguous.
Size the field and collect useful poses
Choose a target that distributes resolvable features over the required field. Preserve the border needed by the detector. For standard perspective-camera calibration with a planar target, capture a variety of positions and tilts while keeping focus and lens settings consistent. Telecentric and specialised optics may require a different model and capture procedure.
- Check image edges, where distortion and reduced contrast may be more apparent.
- Avoid saturated clear areas and specular glare that change detected feature boundaries.
- Keep the target flat in its working mount; a planar model cannot absorb every physical deformation correctly.
Validate with independent lengths and positions
Inspect residuals by field position, then measure a known distance or point set that was not used to fit the model. Low reprojection error shows how well the model fits the input observations; it does not independently establish traceable dimensional accuracy. For tilted dot targets, the centre of a fitted ellipse is not generally the perspective projection of the physical circle centre; use a method appropriate to the required accuracy.