Deciding whether a measurement needs a telecentric lens, and choosing one

In short

Estimate the size change a regular lens gives when a part sits off its calibration plane: length × height offset ÷ (working distance − focal length). If that is small against your measurement budget, calibrate a regular lens and keep it if parts at the highest and lowest positions measure within budget. If the estimate alone uses up the budget, or an angled view hides an edge that a straight-on view would show, plan a telecentric lens. Pick it by magnification: the field, sensor size divided by magnification, must hold the part and its placement spread, and the lens depth must cover your height range.

Why a regular lens misreads parts that vary in height

A regular fixed-focal lens, often called an FA (factory automation) lens, is entocentric: it sees in perspective, so a nearer part looks larger. Calibration sets the millimetres per pixel at one height. When the height changes, the scale changes too, and the same diameter reads differently.

Our calculators use a thin-lens estimate of that change:

|ΔL| ≈ L × |Δz| / (WD − f)

L is the length you measure, and f is the focal length. Δz is how far the part sits from the calibration plane along the lens axis, so it equals the whole height spread only when you calibrate on the highest or lowest part. WD is the working distance, counted from the lens principal plane as choosing a lens when the working distance is fixed explains. The result is a sensitivity, how strongly the reading reacts to height, not a measured error. Positions follow the same rule: the scale changes around the lens axis, so an edge near the side of the field moves more than one near the centre.

Perspective also shows the sides of tall features. Away from the image centre, a regular lens looks slightly into a hole or along a wall, and the wall can hide the edge you measure.

When a regular lens is still enough

A regular lens is not ruled out for measurement. Compare the estimate with your measurement budget, the share of the part tolerance you allow the vision system. If the estimate is small, calibrate at the reference height and measure real parts at the highest and lowest positions. Keep the lens if the calibrated error stays inside the budget, and record the budget and calibration you used. A fixture that presses the part against a fixed reference face shrinks Δz, and the estimate with it.

What a telecentric lens fixes, and what it does not

An object-side telecentric lens keeps its chief rays, the rays through the centre of its aperture, nearly parallel on the part side. Within its specified range, height then barely changes the scale. The lens also views holes and walls nearly straight on, so an edge that an angled view hid can come back into view. The remaining tilt is the telecentricity angle, a datasheet value that is not zero.

A bi-telecentric lens keeps the chief rays nearly parallel on the sensor side as well. Opto Engineering describes two effects of this, compared with a lens telecentric on the part side only: the image of a point can keep a more even shape across the sensor, and magnification can change less when the part moves out of focus. That describes the design rather than a figure for a given lens, so judge a bi-telecentric lens by its datasheet and a height test, like any other.

Neither type removes distortion, the change of scale across the field, or the need for calibration. Neither shows an edge that the part itself hides, such as one under an overhang.

What to compare on a telecentric lens datasheet

• Magnification is the image size on the sensor divided by the object size. The field on each axis is the sensor size divided by it, and one pixel covers the pixel pitch divided by it on the part. Both lenses here have one specified value, while some telecentric lenses offer variable magnification (Edmund Optics), so use the value for your configuration.
• Working distance is set by the lens, so you build the machine to it: MVL-MY-05-110C-MP states 110 ± 2 mm (datasheet). Take the reference face from the maker's drawing, and check clearance for the light and the fixture.
• Depth comes in two kinds. Depth of field is the range that stays sharp enough at a stated aperture and criterion. Telecentric depth is the range over which the stated magnification or measurement criterion holds. A telecentric lens's depth depends on magnification, working F-number, wavelength, pixel size and the edge algorithm (Opto Engineering), so a figure means only what its criterion says. Test it as in deciding whether depth of field covers every part position.
• Telecentricity angle tells you how much perspective is left. Take it from the full datasheet of the exact part number; a complete one, such as Edmund Optics' for its 0.20× lens, also states the criteria behind its MTF and depth figures.
• Image circle, the round image area the lens is specified to project, must cover the active sensor diagonal, as checking whether a lens fits your camera explains; otherwise the circle clips the field that the sensor width promised (Opto Engineering).
• Object-side resolution has two parts. Sampling, the pixels per millimetre on the part, is arithmetic; what the lens resolves comes from its resolution figure or MTF (modulation transfer function) curve, the contrast kept at each detail size. Neither is measurement accuracy.
• Coaxial illumination, light sent along the lens axis, separates a flat reflective face from recesses, which turn dark (KEYENCE). It is an option for such surfaces, not a requirement.

Worked example

Say you measure a hole about 10 mm across, and the vision system may use ±0.05 mm as its measurement budget. You calibrate on the lowest part, and the highest part sits 1 mm closer to the lens. All three values are assumptions of this example.

Is a regular lens enough? Take a regular 25 mm lens with the calibration plane at a working distance of 200 mm (assumption). The formula gives 10 × 1 / (200 − 25), about 57 µm (calculated, thin-lens estimate). By this estimate, perspective alone takes more than the 0.05 mm the budget allows. If a fixture held every part within 0.1 mm of the calibration plane, the estimate would drop to about 5.7 µm (calculated), and a calibrated regular lens would be worth testing. Here the part cannot be held that way (assumption), so plan a telecentric lens.

Field and scale. Pairing A puts the 0.5× MVL-MY-05-110C-MP on the MV-CA032-10GM camera. Pairing B puts the 0.158× bi-telecentric MVL-MBT-0158-178 on the MV-CA050-12UM. The numbers are planned from the datasheet magnification; the installed lens is checked in the acceptance test below.

What to do with it. Pairing A gives the finer scale: the 0.05 mm budget spans about 7.2 px. Its field is only 10.6 mm high, though, which leaves about 0.3 mm on each side of the hole (calculated), so the fixture must place every part that closely. Pairing B has room to spare, but the budget spans only about 2.3 px. Use these counts to rank the pairings rather than to accept one; how many pixels a feature needs explains how sampling relates to accuracy.

Both depth figures are larger than the 1 mm height range (datasheet), so focus over that range looks feasible on paper. They cannot be ranked, though: the 0.5× lens entry lists F9.3 next to 2.98 mm without naming the criterion, and the 0.158× entry lists F4.9 but states ±17.6 mm at F11. Get the criterion behind each figure before choosing on depth.

Pairing B clears the sensor diagonal by only 0.39 mm. That covers the sensor on paper; check corner images from the exact build. The 0.5× lens is marked coaxial: yes (datasheet), so test coaxial light if the face around the hole is flat and reflective.

Accepting a pairing. Calibrate at the reference height, then measure the same real parts at the lowest and highest positions. Accept the pairing if the calibrated error and repeatability stay within the ±0.05 mm budget at both heights, with the edge clearly visible. This test also shows how much the scale drifts over the 1 mm, which the depth figures leave open.

Calculate for your case

For your own lens, work the table out by hand. Divide the active sensor width and height by the magnification for the field, and the pixel pitch by it for the part covered by one pixel; pixels per millimetre is the pixel count divided by the field. Do the perspective estimate and the image-circle subtraction by hand as well.

To reproduce the table's magnification and sampling, open our tools API (application programming interface) query for pairing A or pairing B: values.magnification, values.pxPerMm and values.objectPixelUm match it. The field-of-view calculator links for pairing A and pairing B answer the regular-lens question instead: the page selects a 35 mm lens for A and a 20 mm lens for B, and the field, sampling and verdict it shows, about 15.1 × 11.4 mm at 135.3 px/mm and 66.7 × 55.8 mm at 36.7 px/mm (calculated by the page), belong to that regular lens.

Equipment that fits this example

MV-CA032-10GM and MV-CA050-12UM set the sensor sizes of the example. MVL-MY-05-110C-MP belongs to the 2/3-inch telecentric family, and MVL-MBT-0158-178 to the bi-telecentric family. Other variants sit in the compact and large-format families. Check the mount on the camera and lens pages before pairing them.

Common mistakes

• The field is planned from the sensor width alone, and the image corners come out dark or soft. Compare the circle with the active diagonal, and check corner images when the margin is as small as pairing B's 0.39 mm.
• Lenses are ranked by their depth figures, 2.98 mm against ±17.6 mm, and parts at the ends of the height range measure differently from those in the middle. Compare depth at each build's working aperture under a matching quality criterion, weighing the light and resolution that aperture leaves, and test the scale at both height extremes.
• The 6.9 µm pixel is quoted as system accuracy. Calibrate on a target at the reference height, and report the error left over the full height range.

When this rule breaks

The formula is a thin-lens approximation for small height changes. Working out the thin-lens field exactly at both planes gives about 57.5 µm for the 1 mm case, against 57.1 µm from the formula (calculated). Either way, only a calibrated test on real parts gives the station's error.

Telecentricity holds only over the range the maker specifies, so outside the telecentric depth the stated magnification is no longer promised.

A larger part needs a lower magnification, and so fewer pixels per millimetre on the same camera. If no acceptable telecentric field holds the part, go back to a calibrated regular lens and a fixture that controls height.

Next step

Send us the camera model, the feature and its size, the measurement budget, the height range with your calibration height, and the space for the lens. We estimate the perspective change for a regular lens and work out field and scale for telecentric candidates. From the full datasheets of those part numbers we confirm the working-distance reference face, the depth criterion and its aperture, the telecentricity angle, the image circle and the resolution. If a calibrated regular lens will do, we say so.

Related

• Choosing a lens when the working distance is fixed
• Checking whether a lens fits your camera: image circle, sensor diagonal and mount
• Deciding whether depth of field covers every part position
• Deciding how many pixels a feature needs before you choose a camera

How this material was prepared

Prepared by MVisionPro from the stated sources and the MVisionPro calculation engine. A physical test is claimed only when the material says so explicitly. Read the editorial method.

Editorial status: verified. Content updated 2026-09-25.