Know what this target measures
A concentric-square target is a known length and edge reference. It can establish scale at a defined plane, provide a centre for alignment and reveal nonuniform edge geometry. It does not by itself solve camera intrinsics, produce a resolution rating or certify a complete distortion model.
Three layouts are often described with similar words: solid square-in-square patterns, hollow box-in-box overlay marks and nested frames used for image-scale checks. Confirm which physical geometry is supplied before choosing an analysis method.
Define the geometry before taking a reading
State the nominal side lengths, gaps, line widths, substrate outline and the edge convention. Half the difference between corresponding square widths gives the offset of those boundaries, not necessarily the clear gap. For outlined frames with centreline widths Wouter and Winner and stroke widths touter and tinner, the clear gap is (Wouter − Winner − touter − tinner) / 2.
- Keep all object coordinates in one unit. A drawing dimension is not automatically a measured calibration value; record whether the result uses nominal or inspected coordinates.
- A square has four-fold rotational symmetry. Add a fiducial, corner mark or fixture datum when the X/Y direction must be unique.
Lock the complete imaging configuration
µm/px is a property of the objective, tube optics, adapter, camera sampling, image mode and measurement plane together. Lock the configuration used for the real measurement before recording a scale.
| Change | Practical effect |
|---|---|
| Objective, adapter or zoom | Magnification changes; verify the scale again. |
| 2 × 2 binning | Each output pixel represents a larger sensor area; verify the new scale. |
| ROI crop | Usually changes field coverage, not physical sampling; confirm the pipeline. |
| Digital zoom | Changes display or resampling, not the physical sampling; do not use it as magnification. |
Calculate a scale with a long visible span
Measure a known edge span in pixels and divide the physical length by that pixel distance. Select the longest clean span within the field so endpoint localisation contributes less relative uncertainty. Use the same plane and optical settings as the intended measurement.
Use edge or line-centre detection consistently. A chrome edge is affected by diffraction, blur, threshold and illumination; changing the threshold can shift the reported boundary even when the target is unchanged.
Use multiple frames to estimate a centre
When the centre is the datum, fit several frame edges or corners together and inspect their centre spread. Under perspective, averaging edge or corner coordinates does not necessarily recover the projected physical centre. A square’s diagonal intersection preserves its centre under an ideal planar projective mapping; lens distortion and edge error still need attention. Compare frames using a model appropriate to the view and weight observations by their uncertainty.
The target’s square symmetry does not resolve orientation. If a camera or robot needs a unique coordinate direction, use a fiducial or a fixture datum and record the observed side of the glass.
Separate perspective from lens distortion
A flat target tilted relative to the optical axis can appear trapezoidal and can make opposite spans differ even when the lens has no relevant distortion. Level the target and place it in the intended plane before interpreting edge curvature or spacing.
A nested-square reading can flag a problem or compare setups. A full distortion result needs a defined model, field coverage, coordinate convention and reporting metric such as the selected TV-distortion definition. Do not mix EBU, SMIA or local geometric definitions without naming the metric.
Control illumination, polarity and mounting
Define the clear and opaque regions, then verify their observed contrast under the chosen lighting and detector settings. Reflection does not necessarily produce a simple reversal of transmission polarity. Keep the image unsaturated and use diffuse illumination where glare could move an edge. Support the glass without bowing the active area.
- For a thin glass plate, clamp pressure, bonding and temperature gradients can change the working plane. Inspect the installed condition, not only the loose part.
- A chrome-on-glass target intended for transmission may be unsuitable for reflected-light contrast if the surface produces glare. Specify the observed side and lighting geometry.
Separate dimensional tolerance from measurement uncertainty
For scale s = L / N, independent standard uncertainties in physical length L and pixel span N give approximately u(s)/s = √[(u(L)/L)² + (u(N)/N)²]. If two endpoints have independent equal uncertainty σ pixels, their span uncertainty is √2σ. Correlations and systematic edge bias need separate treatment.
Use a long span only where one scale factor represents the region adequately. A long span across a distorted field gives an average scale and can conceal local variation. Compare horizontal and vertical spans and several field positions before applying a global conversion.
Apply thermal expansion to the measured span
For a uniform temperature change, ΔL ≈ L α ΔT. Use the length being measured and the coefficient for the actual glass grade. The substrate outline is relevant to mounting, while the selected pattern span controls the scale calculation.
Repeat, record and validate
Repeat the scale measurement at several image positions and retain the original captures. Record mean, spread, edge method, temperature and optical settings. Validate with an independent length or stage reference when the result controls a production measurement.
After changing optics, camera mode, binning or the mount, treat the old scale as unverified. A target can be dimensionally stable while the image configuration changes.