Measuring a foot from a photo: why 0.1 mm needs structured light

A custom insole shop wants to measure a customer's foot from a photo, with error below 0.1 mm. A single photograph cannot do that, and no image processing will close the gap: the geometry of one view limits even a steady shot to about 1-3 mm on an adult foot. The fix is not a better camera but a different measurement principle — calibrated structured light or laser profiling, the same triangulation industrial 3D sensors use. This page works through the pixel arithmetic that shows what one photo resolves, the three error terms that set the 1-3 mm figure, the triangulation math behind tenth-millimetre depth, and the parts that turn the idea into a working scanner.

What one photo actually resolves

A photograph yields a length only after you multiply pixels by a scale, and both halves carry error. The pixel half is the easy one, and the pinhole model gives it directly:

Take a typical 12 MP phone main camera: 4,000 pixels across a 5.6 mm sensor, true focal length near 4.2 mm (the 26 mm-equivalent figure divided by the sensor's crop factor).

• Cradled over glass at 300 mm working distance: FOV = 300 × 5.6/4.2 = 400 mm, so the pitch is 400/4,000 = 0.10 mm per pixel.
• Hand-held standing over the customer's foot at 900 mm: FOV = 900 × 5.6/4.2 = 1,200 mm, and the pitch rises to 1,200/4,000 = 0.30 mm per pixel.

The pixels alone do not forbid the tenth-millimetre target — the cradle case sits right on it. But pixel pitch is the floor under every other error, never the accuracy itself: the scale, the geometry and the object each add terms on top. The same arithmetic, generalized to a lens of known magnification, is the object-space pixel size calculation, and it is what makes single-image size measurement work at all.

The three errors that set the 1-3 mm limit

Tilt. One view measures the projection of the foot onto the sensor plane. Tilt the foot or the camera by θ and a true length L reads L·cos θ. On a 260 mm foot, 5° of tilt costs 260 × (1 − cos 5°) = 260 × 0.0038 = 1.0 mm. Even 2° of careful hand-held aim costs 0.16 mm — more than the entire hundred-micrometre budget.

Scale plane. The ruler or glass you calibrate against sits in one plane; the foot's landmark points sit above it. Magnification goes as 1/(WD − h), so each millimetre of height offset changes the scale by about 1/300 = 0.33% at a 300 mm working distance. A 3 mm sock sole exaggerates the length by 300/297 = 1.0101, which is +1.0%, or 2.6 mm on 260 mm. A calibrated camera can replace the ruler — see measuring planar distances with a calibrated camera — but the plane problem stays.

Edges and tissue. The silhouette edge of a soft, curved object cannot be placed better than a pixel or two: ±1 px per end is ±0.2 mm at the cradle's pixel pitch, ±0.6 mm at the hand-held one. Worse, the outline you see from above is the instep, not the sole outline that defines shoe fit, and load visibly changes a foot's length and width.

Stack the terms and the hand-held shot lands at 1-3 mm on an adult foot even when you hold steady over a ruler on glass; a fixed cradle with a bare foot gets to roughly ±1 mm. Now set that against the decision it feeds: one EU shoe size is a 6.7 mm length step (US sizes step 1/3 in = 8.5 mm; a half size is 3.3 mm in both). The photo picks a full size with margin and a half size by luck — and it says nothing about instep height, which is what an insole actually needs.

How structured light reaches a tenth of a millimetre

A structured-light scanner replaces the silhouette with triangulation. A projector throws a coded pattern of stripes or dots across the foot; a camera displaced by a baseline B watches the pattern deform; and every pixel gets its own depth from the triangulation angle. Nothing depends on the operator's aim, and the instep arrives as data, not as a guess.

Depth precision follows the stereo equation:

Worked example: a 5 MP camera, 2,448 px wide on a 2/3" sensor (8.8 mm), with a 16 mm lens gives f = 16 × 2,448/8.8 = 4,451 px. Set the rig at z = 600 mm with a 200 mm baseline and sub-pixel matching to δd = 0.3 px:

δz = 600² × 0.3 / (200 × 4,451) = 108,000 / 890,200 = 0.12 mm.

Halve the working distance and the error quarters (0.03 mm at 300 mm); double the baseline and it halves. Read that as precision — repeatability from shot to shot. Absolute accuracy adds the calibration error, which is why a production rig is calibrated on a plate artifact and then checked on a part of known length. Specification sheets for bench-top structured-light systems typically claim 0.05-0.1 mm point precision, and published foot-measurement studies treat sub-millimetre as the static target: a 2025 lightweight visual method reported up to 1.5 mm accuracy, while motion studies aim below 1 mm to match what static foot scanners already achieve.

Length comes from the same pixels. At z = 600 mm the field is 8.8/16 × 600 = 330 mm wide, so the 5 MP camera resolves 330/2,448 = 0.135 mm per pixel on the foot. Holding a tenth-millimetre pixel pitch across a 320 mm field takes 3,200 pixels across the image; half that pitch (0.05 mm) takes 6,400:

For the choice among the depth families — stereo, ToF, structured light — the industrial 3D camera selection guide compares them on exactly this kind of decision, and the stereo edge accuracy analysis shows why depth edges soften as distance grows.

Which approach fits an insole business

The values assume a fixed rig at the working distances used above; hand-held multiplies them several-fold. The photogrammetry row is the honest version of the many-photos idea: with fixed cameras and coded targets it converges on the structured-light rows, minus the pattern that makes a texture-poor instep matchable. Whichever rig wins, the depth-map-to-point-cloud deprojection is the math underneath it, and height-difference measurement on the resulting cloud is what turns the instep into insole geometry.

Decision per case:

• Size-selection kiosk: cradled photo, bare foot on glass — ±1 mm against a 6.7 mm size step.
• Custom insoles and modified lasts: projected structured-light camera over a platform — a tenth to half a millimetre on length and the full instep volume in one capture.
• Orthotic-grade topography or audit-grade repeatability: laser profiler swept on an axis.

Parts that do the job

One-shot scanner

A projected structured-light 3D camera is the single-box version of the geometry above — projector, camera and triangulation in one housing, one capture per foot.

Mount it over a platform, put the customer's bare foot on glass, and the whole error budget lives in the depth equation instead of the operator's aim.

Laser lines without a motion axis

A laser galvanometer stereo camera steers its beam with mirrors, so it scans a static foot with no axis to build, and the stereo pair pins the geometry.

The topography route

A 3D laser profile sensor fires one profile per trigger and delivers the sharpest z detail of the three.

Sweep it along a 300-400 mm stroke axis to cover a 260 mm foot; the axis encoder supplies the lateral coordinate and the profile precision carries the height. Before mounting anything, read the tilted laser point cloud calibration checks — mounting angle is the classic cause of a tilted scan.

Lens for a hand-built rig

Distortion is the enemy of length measurement. The LS-FFL-0567 is a 16 mm C-mount lens rated for 20 MP on a 1.1" sensor, F2.8-22, under 0.3% distortion.

Stopped down toward F11 it holds focus across the instep's height range, and the lens finder sorts the alternatives by focal length and image circle. Camera bodies for a DIY rig come from the area scan cameras page; complete triangulation systems from the 3D sensors page.

Checks before you commit

• Calibrate on a flat plate artifact, then re-verify length on a gauge block or ball bar of known size; the depth-to-real-distance calibration procedure applies to any triangulation device.
• Scan a test foot ten times and report the spread. If a structured-light rig drifts past ±0.2 mm, look for saturated pattern pixels first — blown highlights on the stripes are the usual culprit.
• Keep the glass plate in the fixture: it defines the sole plane and carries the landmarks.
• Follow the 3D camera working distance and positioning guide when setting height: cover the foot plus about 20 mm of margin on every side.
• Measure bare feet, always; if the customer must keep a sock on, log its thickness with the scan.

When the photo is the right answer

If the shop only needs to pick a full EU size, the cradled photo at ±1 mm beats a measuring tape and costs a phone cradle, a glass plate and discipline. The moment the product is a custom insole, a modified last, or any claim better than half a size, the photo is out of its depth — budget the structured-light rig instead, because the hundred-micrometre target lives in triangulation, not in megapixels.

Ready to spec the scanner? Start the configurator at the camera step:

If you would rather talk the error budget through before ordering parts, the contact page is open to everyone.