M12 lens IR cut filter: is it in the glass or on the board

An M12 board camera that should see your infrared lamp often delivers a black frame, and the first instinct is to order a different lens. The instinct is usually wrong. In this class of hardware the infrared cut almost always lives in a small coated glass plate sitting just in front of the sensor, or in a motorised version of that same plate on the camera board — the lens is innocent more often than not. This guide shows you where the filter physically sits, how to test what a lens really transmits in the near infrared, what moves optically when you take the plate out, and how to split an exposure budget between 850 nm and 940 nm with arithmetic you can repeat on the bench. It is written for anyone building near-infrared inspection or night capture on M12 board-level cameras.

Where does the IR cut live in an M12 lens setup?

In nearly every M12 setup the cut is a separate coated glass plate just in front of the sensor, or a motorised version of that plate on the camera board — the lens barrel itself carries no filtering coating.

There are three places the cut can sit, in the order you will meet them:

1. A fixed coated plate on the sensor board. A bare glass disc, typically about 6.5 mm across and 1.0 mm thick on small sensors, clipped or taped above the sensor's cover glass. Most colour board camera modules carry one.
2. A motorised switchable filter. The same plate on a small hinge, flipped out of the optical path for night mode. Consumer day/night boards drive it from a photoresistor; industrial builds drive the control input from a controller output, so the state is commanded rather than sensed.
3. A coating in the lens itself. The exception. A minority of security-line M12 lenses do filter in their own glass, which is why the lamp test later in this guide exists.

That order has a commercial consequence: 'visible only' on a lens data sheet usually means the maker does not specify transmission past roughly 700 nm, not that the glass blocks it. The plate does the blocking, so an ordinary M12 lens and an 850 nm lamp are usually compatible once the plate is out of the path. Mechanically, M12 is a 12 mm thread at a 0.5 mm pitch, and because focus lives and dies by that thread, M12 lens thread pitch and focus shift: the physics of S-mount focus covers its behaviour; the lenses topic page collects the rest of the S-mount guides.

Why does silicon see near infrared at all?

Silicon photodiodes respond to photons out to roughly 1,100 nm, so a sensor keeps collecting near infrared whether you want it or not; the filter in front of it, not the lens, decides how much gets through.

A silicon pixel converts every photon whose energy exceeds the material's bandgap — about 1.1 eV, a wavelength near 1,100 nm. The band from 750 to 1,000 nm sits comfortably inside that window, so an unfiltered sensor is already a competent near-infrared camera, and nothing about the lens changes that.

Two consequences follow. Quantum efficiency falls as wavelength climbs, so the longer wavelength delivers materially less signal than the shorter one — numbers in the lamp section below. And on a colour sensor the Bayer dyes pass NIR unequally: the red channel collects the most, green less, blue least. That asymmetry is the pink-red cast in every unfiltered colour image, and auto white balance cannot fully remove it because the information is genuinely lost. A mono sensor has no dye array, so every pixel collects whatever arrives.

For the colour-specific trade-offs — which filter option to keep and when — Color camera near infrared imaging: IR-cut filter options is the companion read; this guide stays on the physical filter itself.

What does the filter plate on the sensor actually do?

A 650 nm cutoff plate passes visible light and blocks wavelengths longer than about 650 nm, which removes the red haze a colour sensor would otherwise record and restores accurate colour; mono sensors use the same plate to keep stray daylight NIR out of the exposure.

The plate is optical glass with a hard interference coating that transmits the visible band and blocks the band above its cutoff — 650 nm is the common value on small-format plates. It sits close to the sensor so it fits inside the shallow lens seat and stays out of the dust-prone air in front of the lens. The coating decides which wavelengths pass; the glass decides where focus lands.

Any plane-parallel plate in a converging beam moves the focal plane away from the lens, by a predictable amount:

For a typical 1.0 mm plate in glass of index n ≈ 1.5: Δ = 1.0 mm × (1 − 1/1.5) = 1.0 × 0.333 = 0.33 mm. Remove the plate and the best-focus plane moves 0.33 mm toward the lens, so you screw the lens in by 0.33 / 0.5 = 0.66 of a turn — about 240° on the M12×0.5 thread. Refocus from there under your real lamp, then lock it as M12 lens focusing: adjust, check corners and lock describes.

If you want the layer underneath — substrate glass types, coating stacks, how cutoff steepness is specified — the IR cut-off filter glass selection notes for CMOS machine vision cover that half of the decision.

How do you test whether your M12 lens passes 850 nm?

Point the camera at an 850 nm LED lamp with the plate out of the path: bright grey on the lit target means the lens transmits NIR, while a near-black frame means the glass is absorbing it or the plate never left.

Ten minutes and three parts settle it:

1. Take the filter out of the path. Peel the fixed plate off its holder, or drive the switchable one to its open position, and confirm you can see the sensor's cover glass.
2. Light a target at your real wavelength. An 850 nm LED flood is the honest source; for a rough check at the longer wavelength, a household remote control's LED emits near 940 nm — press and hold any button.
3. Shoot lamp-off and lamp-on at fixed exposure and record the mean grey level of the lit area.
4. Read the result. Bright grey with the lamp on: the lens transmits NIR. Near-black: the glass is absorbing it, or the plate is still in place.

Turn the grey levels into a budget number. If white light needs 4 ms for mid-grey and the infrared lamp needs 12 ms through the same lens, the transmission penalty is 12 / 4 = 3×, and that factor goes straight into your exposure plan. Expect a worse figure than white light even from a lens that passes well, because visible-tuned coatings transmit less out of band. If the test fails, replace the lens rather than argue with it — the M12 board lenses page filters by type, and the low-light designs are the right place to start.

What changes optically when you remove the plate?

Removing the plate buys you NIR sensitivity but costs focus accuracy: the same lens focuses 850 nm light at a slightly different distance than visible light, so plan to stop down to F5.6-F8 and refocus under the NIR lamp itself.

Three things move, and only one is permanent.

Focus offset. The 0.33 mm shift from the plate's glass is a one-time correction: screw the lens in about two thirds of a turn and it is closed. How that offset interacts with thread travel and back focus on these boards is covered in Lens focal length vs back focus in board cameras.

Chromatic focal shift. Glass bends 850 nm light slightly differently from visible light, so even after the offset fix the NIR focal plane can sit tens of microns from your white-light focus. Refocus under the lamp, then stop down to F5.6-F8 so depth of field covers the residual error. Depth of field scales roughly linearly with f-number, so each stop you close doubles it — work your exact numbers in the depth-of-field calculator:

Light budget. Stopping down costs signal: F2.8 to F5.6 is two stops, a 4× loss. The case for staying wide — signal when you are photon-starved — is made in M12 lens light gathering: pick a lower f-number, same focal length; in single-wavelength NIR work, depth of field usually wins, because a sharp F5.6 frame beats a brighter, half-blurred F2.8 one. Geometry itself barely moves: refocused, the field of view you laid out in white light stays your field of view.

Should you pick an 850 nm or a 940 nm lamp?

Choose 850 nm when sensor signal matters most — NIR-enhanced sensors publish quantum efficiency around 60% at 850 nm against 40% at 940 nm — and choose 940 nm when the lamp must stay invisible, paying roughly 1.5× the exposure for the same lamp power.

Both wavelengths sit inside silicon's response and pass the same lenses; they differ in signal strength and in visibility. Conventional sensors fall further than the enhanced figures, most of all at the longer wavelength. The shorter-wavelength LED also shows the eye a faint red point, while the longer one is effectively invisible — which matters near people, or when the inspection should not advertise itself.

A worked exposure budget. Your visible setup exposes 4 ms at F2.8; you move to a 940 nm lamp of equal optical power.

• QE penalty: 4 ms × (60 / 40) = 6 ms.
• Refocus and stop to F5.6 for the focal shift: light falls by (5.6 / 2.8)² = 4, so 6 ms × 4 = 24 ms.
• Budget check: at 30 fps the frame lasts 33.3 ms, so 24 ms fits with about 9 ms to spare.
• Push to F8 for depth of field and the factor becomes (8 / 2.8)² ≈ 8.2: 6 ms × 8.2 ≈ 49 ms, over budget. Drop to 20 fps (50 ms) or raise lamp power instead.

Decision: the shorter wavelength when exposure is tight or the lamp must be small; the longer one when invisibility is a requirement and you can pay the 1.5× plus a stop or two. For the illumination geometry — flood angle, working distance, projector alternatives — Infrared flood illumination for stereo depth imaging works through the same trade space.

Which parts make NIR imaging with M12 optics easy?

Pick a mono, global-shutter M12 board camera, a lens whose type line already expects low light, and a lamp at the wavelength you tested, because a mono sensor has no Bayer colour array and every pixel then collects the NIR signal.

A mono sensor is the single biggest lever; every board camera listed as Colour/Mono has that variant, and it is the one to order. Global shutter covers anything that moves during exposure, and a low-light-class lens type keeps the transmission curve friendly out where you are working.

The MV-CB016-10GM-M-W is the compact fit: mono, global shutter, M12 mount, GigE output, 1/2.9" sensor. The LS-BCM-0061 is the same idea in a miniature-industrial package: 1.3 MP over GigE on a 1/2.7" sensor.

Two lens lines cover the common working distances:

The LS-MBL-0198 is a 4 mm starlight low-light lens sized for 1/2.7" sensors — the pick when the camera sits close to the part. The LS-MBL-0113 is a 12 mm F2.8 lens covering a 1/1.8" sensor with an 8-11 mm image circle, for stands that need working room.

Before you order, close four checks:

• the lens image circle covers the sensor format, as an 8-11 mm circle does a 1/1.8" sensor;
• the camera is the mono variant, not colour;
• you know how the filter stays out of the path in production — plate removed and recorded, or switchable input wired to an output you command;
• the lamp wavelength matches the wavelength you tested, because a lens that passes the shorter wavelength can still fall away at the longer one.

Shortlist on the board-level cameras page, where the filters split mono, shutter and mount in one pass. M12 lens selection for board cameras: format, focal length and IR cut is the companion read for the wider choice, Board camera evaluation checklist for inspection systems walks the bench checks, and when the shortlist is down to two, the lens configurator pairs either one to your sensor.

When is the usual answer wrong?

The plate is not the only thing that can block your lamp: a minority of M12 lenses cut infrared in their glass, a switchable board can flip its filter back in on photoresistor logic, and an unfiltered colour camera loses more colour accuracy than the NIR signal gains.

Four cases break the pattern, in the order you are likely to meet them.

The lens itself blocks NIR. Those security-line M12 lenses carry filtering in their own glass, and no plate removal fixes that. The lamp test is the referee: a black frame with the plate out means replace the lens, not the plan.

The board switches the filter back in. Boards with a photoresistor decide filter state from ambient light, which inside a lit factory means the filter is always in. Take the decision away from it: disconnect the auto input, or drive the filter's control wires — a ground pin and a switching input — from a spare GPIO or controller output, so the state at the moment of exposure is commanded and repeatable.

You need colour accuracy. Removing the plate buys NIR and destroys colour fidelity; no software step recovers colour from a sensor drenched in infrared. Keep the plate for colour work and add a mono camera for the NIR channel — two cameras, one trigger, two clean images.

The longer wavelength starves a conventional sensor. If the sensor's quantum efficiency there is a small fraction of its value at 850 nm, no exposure juggling rescues the frame; move the lamp to the shorter wavelength or specify an NIR-enhanced sensor.

So the decisions, per case: NIR-only inspection — remove the plate, run mono, lamp at the shorter wavelength, refocus, stop down a stop or two. One camera for day and night — buy the switchable filter and command it from your controller. Colour-accurate work — leave the filter in and add a mono channel. Black frame after the plate test — change the lens. The camera finder filters by mono, shutter and mount, and the contact page team will check a lamp-lens-sensor combination against your working distance before you order.