Bandpass Filters and IR-Cut Filters for Machine Vision: CWL, FWHM, and Day/Night Architectures
IR-cut, bandpass, dual-bandpass, and no-filter architectures for machine vision cameras: when each applies, how CWL, FWHM, and OD blocking work, and how mechanical switching fits in.
Four filter architectures cover almost every machine vision imaging mode. An IR-cut filter blocks NIR above roughly 650nm to protect visible-color accuracy. A bandpass filter passes one narrow band, such as 850nm ±25nm, to isolate an active illumination wavelength from ambient light. A dual-bandpass filter passes visible light plus one NIR band simultaneously, giving passive day/night operation with no moving parts. Running no filter at all passes everything the sensor can detect, which is correct when the system is NIR-only or when the sensor handles spectral separation itself (RGBIR, multispectral). Commonlands encodes the choice per lens through the ir_cut_off_filter variant (650/660nm IRC vs NIR), and browsing the filters collection shows every standalone option.
Which filter does your system need?
Two questions decide the filter architecture: what does the sensor need to capture, and where does the light come from? If the output must be an accurate visible-color image under ambient or white light, the system needs NIR blocked. If the system is illuminated by an 850nm or 940nm LED and needs to reject ambient light around that band, it needs a narrow passband instead. If one camera must do both, at different times or simultaneously, a passive or active dual-mode architecture is required.
| Architecture | Passes | Best use case | Main tradeoff |
|---|---|---|---|
| IR-cut filter | Visible only, blocks NIR above ~650nm | Color inspection, label reading, any daylight-only system | Attenuates NIR to the filter's OD blocking level, so it is effectively incompatible with IR illumination |
| Bandpass filter | One NIR band, e.g. 850nm ±25nm | Active NIR illumination, barcode reading, presence detection | Blocks visible light, so no color imaging is possible |
| Dual-bandpass filter | Visible + one NIR band, simultaneously | Passive day/night with no moving parts | Some NIR bleeds into daytime color; some ambient bleeds into NIR mode |
| No filter | Full sensor spectral range (~400-1000nm) | Pure NIR inspection, RGBIR sensors, multispectral systems | Daylight color severely degraded if any visible imaging occurs |
| Mechanical IR-cut switcher | Visible in day mode, full NIR in night mode | True day/night where daytime color accuracy is a spec requirement | Moving parts, drive electronics, finite switching lifetime |
The sections below work through each architecture in more depth, then show how Commonlands encodes the choice into lens part numbers via the ir_cut_off_filter metafield. For the underlying illumination-wavelength decision, see NIR imaging machine vision.
A filter does not fix poor focus, an inadequate working distance, motion blur, or specular glare from the illumination source itself. It is one component in an optical stack, not a substitute for correct lens selection, sensor choice, and illumination geometry. See working distance in machine vision lenses and sensor size and lens compatibility for how those decisions interact with the filter choice.
IR cut filters: when to use one and when to skip it
An IR-cut filter blocks near-infrared light above roughly 650nm before it reaches the sensor. Bare silicon photodetectors respond from about 300nm to 1100nm, of which a practical camera module typically retains roughly 400-1000nm after cover glass, microlens, and dye losses, and Bayer-pattern RGB dye filters do not fully separate that range from NIR: a photon at 800nm passes through red, green, and blue filters almost equally, contaminating all three channels at once. Without an IR-cut filter, daylight images shift toward red or pink, white balance drifts, and color classification loses contrast.
Keep the IR-cut filter, or specify the 650/660nm IRC lens variant, whenever the application needs accurate RGB output under ambient or white-light illumination: color inspection, label verification, any classification model trained on visible-spectrum images, or any deployment where sunlight, halogen, or incandescent sources contribute significant NIR energy. Skip it, or specify the NIR (no-filter) variant, when the system deliberately uses 850nm or 940nm illumination for night operation, subsurface defect detection, RGBIR sensing, or multispectral imaging. An IR-cut filter in that path will make the illuminator look dim or completely dark, which is a common integration mistake, not an illuminator fault.
| Factor | With IR-cut filter | Without (NIR variant) |
|---|---|---|
| Daylight color accuracy | Good: only visible light reaches the sensor | Degraded: NIR contaminates the RGB channels |
| NIR sensitivity (850/940nm) | Gesperrt | Full NIR response |
| IR LED-illuminated inspection | Unsuitable | Suitable |
| Low-light photon budget | Reduced: NIR photons are removed | Higher: the full 400-1000nm range is available |
For the cutoff wavelength, 650nm is the standard starting point for daytime color imaging, blocking NIR at the cost of truncating deep red (650-700nm). Some designs use 680nm or 700nm cutoffs instead, which retain more red-channel response at the cost of more NIR bleed. For package format, fixed filters ship in round and square formats sized to the sensor, commonly 7mm round, 7mm square, and 10mm round for M12 systems, with custom sizes available for larger image circles. Integration is either a separate filter holder, a mechanical switcher, or bonding the filter directly to the lens via custom filter gluing (the lowest stack height, but not field-removable).
These solve different problems. An IR-cut filter is a spectral component that blocks NIR. It does not bring visible and NIR to a common focus. An IR-corrected lens is an optical design that brings visible and NIR wavelengths to a common focal plane, so focus does not shift when switching illumination modes. Day/night systems typically need both: the IR-corrected lens holds focus in either mode, and the filter or switcher handles the spectral blocking. See the IR corrected lens machine vision guide.
Filter glass in the path also shifts back focal length slightly. See the angle-of-incidence section below for the related effect on bandpass filters.
Bandpass filters: CWL, FWHM, and OD blocking
A bandpass filter passes a defined wavelength band and attenuates everything outside it, isolating the illumination wavelength a system uses from ambient light and off-wavelength interference. Three parameters define its spectral performance: center wavelength, passband width, and out-of-band blocking.
Center wavelength (CWL)
CWL is the passband midpoint, typically the average of the 50% transmission wavelengths on each side of the peak. Match it to the illuminator's peak emission from the LED datasheet, not the marketing name: many "850nm" LEDs actually peak between 840nm and 870nm depending on temperature and drive current. Most LED illuminators emit with a 20-50nm FWHM themselves, so the filter CWL only needs to fall within that emission band, not sit exactly on the nominal wavelength.
Full width at half maximum (FWHM)
FWHM is the passband width measured at the 50% transmission points. Narrower FWHM gives more selective ambient rejection; wider FWHM tolerates more CWL mismatch and angle-of-incidence shift but passes more ambient. For outdoor NIR systems with broadband ambient, 25-40nm FWHM is a practical starting range: narrow enough to reject sunlight, wide enough to capture a typical LED's own emission spread and tolerate off-axis passband shift.
Out-of-band blocking (OD rating)
Optical density (OD) measures attenuation outside the passband. OD4 corresponds to 10,000× attenuation, or under 0.01% transmission, which is sufficient for most machine vision ambient-rejection tasks including outdoor sunlight. Very bright ambient or high-sensitivity sensor settings may need OD5. Blocking is not always uniform across the full out-of-band range. Check the transmission curve if the sensor has high quantum efficiency far from the passband.
Interference-based bandpass filters, including the current Commonlands bandpass filters, use thin-film coatings rather than absorption dyes, giving high peak transmission (above 90%) and precise spectral edges, but also making performance sensitive to the angle light enters at, which is covered in the angle-of-incidence section below.
Installing an IR-cut filter in a system that also runs 850nm LED illumination and then diagnosing the illuminator as underpowered. The illuminator is working; the filter is blocking it. If the illumination source is NIR, the filter in the optical path has to transmit that wavelength, not block it.
850nm vs 940nm bandpass filters
Both wavelengths work with silicon CMOS sensors and LED illumination; the choice comes down to sensor sensitivity, visible glow, and illuminator power budget. Silicon sensors typically have several times higher quantum efficiency at 850nm than at 940nm, so an 850nm system produces a brighter signal for the same illuminator power and exposure, letting the illuminator run at a lower drive current and cost less. Wall-plug efficiency of the LED itself is not a reliable part of this advantage: modern 850nm (AlGaAs) and 940nm (GaAs) emitters have broadly comparable electrical-to-optical efficiency, and the specific emitters chosen can shift that comparison either way. The drawback is that 850nm LEDs emit a faint visible red glow perceptible at moderate to high drive power. The glow is often irrelevant inside an industrial enclosure but a problem in consumer or human-facing installations.
940nm sits beyond the visible range and is invisible to nearly all human observers and standard color cameras, which is why it wins for covert or human-environment illumination. The tradeoff is meaningfully lower sensor quantum efficiency at 940nm, which typically requires more illuminator power, longer exposure, or higher gain to reach equivalent image quality. This exposure penalty comes from sensor QE, not from LED efficiency (depending on the specific emitters chosen, 940nm LEDs can match or exceed the wall-plug efficiency of 850nm LEDs), so budget illuminator power and thermal headroom against the sensor-side QE gap rather than assuming an additional LED-efficiency penalty stacks on top of it.
Choosing between them
The filter choice follows the illuminator: if the illuminator emits at 850nm, use an 850nm bandpass; if 940nm, use a 940nm bandpass. Neither wavelength is universally superior. The decision starts from visible-glow tolerance and illumination power budget, then works forward to the filter spec. For the CBP940 filter, Commonlands specifies T>90% at 940nm, matching standard 940nm LED illuminators. For the CBP850, the same T>90% specification applies at 850nm. Both are interference filter designs with OD-level blocking outside their passbands.
Before committing to 940nm, verify the illumination power budget and sensor noise floor allow adequate image quality at the required frame rate. The lower QE is not a minor derating; it can mean doubling illuminator drive current or exposure time relative to an 850nm system covering the same working distance. For an illumination-first breakdown of this tradeoff, see the NIR imaging machine vision guide.
IR cut vs bandpass filters: a side-by-side comparison
An IR-cut filter and a bandpass filter do opposite things to the spectrum and are not interchangeable. An IR-cut filter at 650nm transmits visible light and blocks NIR, protecting color accuracy on a broadband-illuminated color sensor. A bandpass filter at 850nm transmits NIR and blocks visible light, isolating an active illumination wavelength from ambient noise. Installing an IR-cut filter in a system that also runs 850nm LED illumination is a common integration mistake. The illuminator is not underpowered; the filter is blocking it. The confusion is understandable since both are thin optical elements sitting in roughly the same position in the stack, but their transmission curves are mirror images of each other by design.
| Factor | IR-cut filter (650nm) | Bandpass filter (e.g. 850nm) |
|---|---|---|
| Primary function | Blocks NIR to protect color | Isolates one band, blocks the rest |
| Best imaging mode | Visible-color under broadband light | Active NIR illumination systems |
| Visible color accuracy | Ausgezeichnet | None: visible light is blocked |
| 850nm sensitivity | None (blocked) | High (passes the band) |
| Ambient rejection | Removes NIR ambient only | Blocks visible and most NIR ambient |
The two questions that settle the choice are what the sensor is capturing and where the light originates. If the sensor is producing a visible-color image under ambient, white LED, or fluorescent illumination, an IR-cut filter at 650nm is correct: the illumination is broadband, color accuracy matters, and NIR contamination from that broadband source is the problem to eliminate. If the sensor is imaging under 850nm LED illumination and needs to reject ambient light, a bandpass filter centered at 850nm is correct: the illumination is narrowband, color imaging is not the goal, and ambient rejection is the problem to solve.
If neither a fixed IR-cut filter nor a fixed bandpass filter cleanly describes the system, because it needs both visible-color and NIR imaging, the architecture changes to a dual-bandpass filter or a mechanical switcher, covered next.
Dual-bandpass day/night filters
A dual-bandpass filter passes two spectral windows through a single fixed element: the visible RGB range, roughly 400-680nm, and a narrow NIR band centered at 850nm or 940nm, blocking the transition region between them (roughly 700-800nm). By day, ambient or artificial visible light passes through for a color image while the 700-800nm band is blocked. At night, the system switches on NIR LED illumination and the filter passes that band for a monochrome image. Mode switching happens entirely by turning the illuminator on or off: the filter has no moving parts, no control signal, and no power requirement.
The daytime tradeoff is that the NIR passband stays open at all times; there is no mechanical switcher to close it. Ambient NIR at 850nm passes through alongside visible light and contributes to sensor output, producing a measurable color shift that is difficult to fully correct in post-processing. Within the 850nm dual-bandpass category, Commonlands offers two transmission levels that address this exact tradeoff: the CDB850 at greater than 90% transmission at 850nm, prioritizing nighttime exposure margin, and the CDB851 at 60% transmission, prioritizing daytime color accuracy at the cost of roughly one-third less nighttime NIR throughput. A 940nm variant, the CDB941, exists for covert installations where 850nm's visible glow is unacceptable.
A single-bandpass NIR filter, such as the 850nm bandpass filter, passes only the NIR band and blocks visible light entirely. It cannot support daytime color imaging. A dual-bandpass filter passes both bands from one element.
A passive dual-bandpass design is the right choice when moving parts are not acceptable due to vibration, shock, or maintenance access, when the module has no physical room for a switcher, or when moderate daytime color shift is acceptable for the application. Outdoor perimeter cameras and general robotics platforms are common fits. Pair a dual-bandpass filter with an IR-corrected M12 lens to hold focus across both visible and NIR; a standard lens shifts focus between the two bands. When daytime color accuracy is a hard specification, evaluate the mechanical switcher covered next instead.
Which transmission level should you choose?
Choose the greater-than-90% variant (CDB850) when nighttime exposure margin is the binding constraint and daytime color shift is acceptable for the application. Choose the 60% variant (CDB851) when daytime color accuracy is the higher priority and the nighttime illumination power budget can absorb the roughly one-third reduction in NIR throughput. Validate both conditions with the actual sensor and lens: measure daytime color balance under realistic outdoor daylight, not studio lighting, and measure nighttime signal-to-noise ratio at the operating illuminator power and sensor gain before finalizing either variant.
IR cut filter switchers: the active day/night architecture
A mechanical IR-cut filter switcher moves an IR-cut filter in and out of the optical path between the lens and sensor on an electromechanical carrier. In day mode, the IR-cut filter sits in the path, blocking NIR for accurate color. In night mode, the carrier moves the filter aside and replaces it with a plain glass dummy window of matched thickness, restoring full NIR sensitivity. Unlike a dual-bandpass filter, the switcher fully removes NIR from the path in day mode: there is no residual passband to cause color contamination, which is the core reason to choose a switcher over a passive dual-bandpass architecture when color accuracy is a specification requirement.
The dummy window matters because a glass element in a converging beam shifts the focal plane. If the IR-cut filter is simply removed for night mode with nothing in its place, the optical path shortens and the image goes soft. A dummy window of matched thickness (the Commonlands CLA214-ICR pairs a 0.21mm IRC filter with a 0.21mm dummy glass) keeps total glass thickness constant across both modes so focus does not shift. Some designs use an intentional thickness offset instead: the CLA216-ICR is a day/night-compensated switcher whose filter and dummy-glass thicknesses differ (a 0.30mm IRC filter is paired with a 0.21mm dummy glass), adding a small amount of focus shift that partially compensates for longitudinal chromatic aberration on lenses that are not IR-corrected, improving NIR focus at the sensor in night mode. Confirm the exact glass thicknesses and the resulting compensation on the switcher datasheet.
A standard switcher's dummy window opens the sensor to the full 700-1000nm NIR range in night mode, including ambient NIR from sources other than the intended illuminator. Where that broad acceptance reduces contrast, a hybrid switcher replaces the dummy window with an 850nm bandpass filter instead. The CLA216-ICR-850BP carries a 650nm IR-cut filter on one side and an 850nm bandpass filter on the other, giving band-selective night imaging rather than broad NIR acceptance. This matters when the deployment has competing NIR sources: other cameras' 850nm illuminators, solar background at dawn or dusk, or NIR-emitting equipment nearby. If the nighttime environment is controlled and indoors with a single known illuminator, the standard dummy-window switcher is usually adequate and less costly.
The drive electronics are simple. Commonlands M12 switchers such as the CLA214-ICR use a 3.3V H-bridge drive to reverse the actuator between its two positions with a short pulse rather than continuous power, so do not apply continuous DC voltage to the coil. Confirm the exact drive voltage, drive-pulse duration, and holding-current behavior on the switcher datasheet. The host system needs two GPIO lines or a small H-bridge driver IC to source the forward and reverse pulses; a single logic-level signal is not sufficient because the actuator needs the polarity reversed to drive it the other direction, not merely switched on and off. Allow a brief settling time after the pulse before capturing a frame, since the mechanical carrier is still completing its travel and settling against its end stop immediately after the drive pulse ends; confirm the required settling window on the datasheet. Commonlands M12 switchers such as the CLA214-ICR are rated for a minimum of 50,000 switching cycles; confirm the rating on each switcher's datasheet. A camera switching twice daily reaches 50,000 cycles in 68 years, but a system switching 100 times per day reaches the limit in 1.4 years, so calculate against the actual deployment duty cycle. A twilight-triggered outdoor camera switching once per dawn and once per dusk sits close to the twice-daily case; a system that toggles on every cloud passing over a light sensor can burn through cycles far faster than the nominal daily count suggests, so debounce the day/night trigger signal rather than switching on every ambient reading. Validate switching repeatability and focus position in both modes across the operating temperature range before finalizing a switcher-dependent design, since actuator response time and latch force both drift with temperature and can affect switching reliability at cold-start conditions.
IR filter vs no filter: the M12A650 / M12ANIR convention
Commonlands encodes the filter decision directly into the lens part number through the ir_cut_off_filter variant. A lens with a 650nm (or 660nm on some product lines) IR-cut filter glued into the filter slot is the IRC variant, appropriate for daylight visible-color imaging. A lens with no filter installed is the NIR variant, passing the full spectral range the sensor can detect, appropriate for NIR imaging, RGBIR sensors, and day/night operation under IR illumination. The optics are identical between variants: only the filter installation differs, and it is bonded, not field-removable, so the choice is made once at order time.
Keep the IRC variant when the application needs RGB color under ambient or white light: color inspection, label verification, or any pipeline where color data feeds a model trained on visible-spectrum images. Choose the NIR variant when the system is built around 850nm or 940nm illumination: night surveillance, subsurface defect detection, or an RGBIR sensor that needs NIR photons reaching its dedicated NIR pixels. A NIR-variant lens used in daylight does not become a general-purpose upgrade; it produces worse color than the IRC variant in the same ambient conditions.
Neither fixed variant handles a camera that must do both color and NIR imaging with full quality in each mode. Pair a NIR-variant lens with a dual-bandpass filter such as the CDB851 for a passive, no-moving-parts compromise, or with an active switcher such as the CLA216-ICR-850BP when both modes must meet full independent specification. Always verify the specific filter designation on the current product page or datasheet before specifying, since P/N conventions can vary slightly by lens family.
Why the filter is not field-swappable
Commonlands bonds the filter to the lens element with an adhesive process rather than seating it in a removable holder. Bonding eliminates a separate filter holder from the mechanical stack, reduces overall stack height, and prevents the filter from shifting or rattling loose in a vibration environment, all real benefits for a fixed-focus embedded camera. The cost is that the filter decision becomes permanent at the time of purchase: an M12A650 lens cannot be converted to NIR use in the field, and an M12ANIR lens cannot be converted to filtered color use, without replacing the lens or adding a separate filter component downstream in the optical path. Confirm the variant against every lighting condition the deployment will encounter, including seasonal daylight variation, before committing to a production order.
Angle-of-incidence shift in compact optics
Interference-based bandpass and dual-bandpass filters perform to spec only near normal incidence. When light enters at an angle, the effective path length through the coating changes, shifting the passband toward shorter wavelengths: typically a few nanometers at 10° angle of incidence (AOI), growing rapidly with angle, and enough at 30° or more that the passband may no longer fully capture the illumination peak. The magnitude depends on the coating design, so confirm the shift against the specific coating's angular-shift data.
In a long-focal-length lens, off-axis chief rays reach the sensor at small angles, so a filter near the sensor sees near-normal incidence across the field. In a short-focal-length or wide-angle M12 lens, off-axis chief rays can reach 20-30° at the periphery, so the same filter sees a range of incidence angles from near-normal at center to steep at the corners. For a 30nm FWHM filter, a 5nm peripheral shift is often tolerable if the illuminator's own emission band is wider than the filter passband; for a 10nm FWHM filter, the same 5nm shift can push the illumination peak to the passband edge, visibly darkening the corners.
Check this by comparing the expected maximum AOI at the sensor periphery against the filter passband width and the illuminator's emission FWHM. If the AOI-induced shift stays within the illuminator's emission width, the system should perform evenly across the field. In the air space between the lens's last surface and the sensor, a chief ray travels in a straight line at a fixed angle, so its angle of incidence on a flat filter is the same at every axial position in that space: moving the filter closer to the lens does not reduce chief-ray AOI. Effective mitigations are a wider FWHM filter more tolerant of shift, a shorter illuminator CWL so the shifted passband still covers the peak at the periphery, or selecting a lens with lower chief ray angle (more image-space telecentric). Any filter glass in the optical path, coated or not, also shifts the effective back focal length by a small amount tied to thickness and refractive index. This is worth checking on lenses positioned close to the sensor with tight chief-ray-angle tolerance, independent of the AOI passband effect.
The practical symptom of an unaddressed AOI shift is corner darkening or a subtle color cast at the edges of the frame that is easy to misattribute to lens vignetting or uneven illumination rather than the filter. Distinguishing the two matters: lens vignetting is usually radially symmetric and roughly constant regardless of illumination wavelength, while an AOI-shift artifact tracks the filter's passband and illuminator spectrum and can often be reduced or eliminated by swapping to a wider-FWHM filter without changing the lens or illuminator geometry at all. Before assuming a lens or illumination layout problem, swap in a wider-FWHM filter of the same CWL and see whether the corner falloff improves; if it does, the AOI shift was the dominant cause, not vignetting.
This effect applies to single-bandpass, dual-bandpass, and the bandpass side of a hybrid switcher alike: any interference coating in a converging beam is subject to the same AOI physics. It is most often overlooked on compact, wide-angle M12 designs where the front element sits close to the sensor and the field of view is large. For how sensor format and lens compatibility relate to chief ray angle at the periphery, see sensor size and lens compatibility and the angle of view calculator.
For focal-length selection tradeoffs that affect chief ray angle at a given sensor format, see how to choose focal length for machine vision.
Top machine vision filters by use case
The right machine vision filter follows the illumination mode. Commonlands stocks the CBP850 and CBP940 single-band NIR filters for active infrared illumination, the CDB850 dual-bandpass filter for passive day/night, the CSP650 IR-cut filter for daylight color accuracy, and the CLA214-ICR mechanical switcher for day/night that must meet a color spec. The table ranks them by how often each resolves a filter decision, with an honest competitor entry for comparison.
| # | Filter | Spectral spec | Primary use case | Link |
|---|---|---|---|---|
| 1 | CBP850 single bandpass | CWL 850nm, T>90% | Active 850nm illumination; highest silicon QE of the NIR options | CBP850 850nm bandpass filter |
| 2 | CBP940 single bandpass | CWL 940nm, T>90% | Covert illumination with no visible red glow | CBP940 940nm bandpass filter |
| 3 | CDB850 dual-bandpass | Visible + 850nm, T>90% NIR, blocks 700-800nm | Passive day/night with no moving parts | CDB850 dual-bandpass filter |
| 4 | CSP650 IR-cut | Shortpass, blocks NIR above ~650nm | Daylight color accuracy on NIR-sensitive CMOS sensors | CSP650 650nm IR-cut filter |
| 5 | CLA214-ICR / CLA216-ICR switcher | 650nm IR-cut moved in and out, 50,000-cycle, 3.3V drive | Active day/night where daytime color is a specification | CLA214-ICR IR-cut filter switcher |
| 6 | MidOpt (Midwest Optical) | Broad off-the-shelf CWL catalog, visible and NIR | Threaded C-mount thread-in filters across many wavelengths; typically higher per-unit cost than bare filter glass | midopt.com |
These are the filters Commonlands stocks against the four architectures covered above, ordered by how frequently each resolves a real machine vision filter decision. Spectral values are the datasheet figures cited elsewhere in this article. MidOpt is listed for landscape completeness: its threaded catalog covers more CWLs off the shelf, at a higher unit cost, and it is not a Commonlands product.
For the illumination-wavelength decision behind these picks, see the NIR imaging machine vision guide, and pair any of them with an IR-corrected M12 lens when focus must hold across both visible and NIR. Product cards with pricing and formats follow below.
A practical filter selection checklist
Work through these steps in order before specifying a filter. Each step narrows the decision and reduces the risk of a mismatch between the filter's specification and how the system actually behaves once deployed with a real sensor and illuminator.
- Define what the sensor must capture. Accurate visible color, isolated NIR, or both at different times.
- Identify the illumination source and its true peak wavelength from the LED datasheet, not the marketing name.
- If visible color only, use a fixed IR-cut filter (650/660nm IRC lens variant or standalone CSP650). Confirm no NIR illumination exists anywhere in the system.
- If NIR only, use a lens with no filter (NIR variant), adding a standalone bandpass filter if ambient or off-band NIR needs rejecting.
- If both are required and color accuracy can flex, use a dual-bandpass filter matched to the illumination wavelength (850nm or 940nm).
- If both are required at full spec, use a mechanical IR-cut switcher, or the hybrid IR-cut/bandpass variant if nighttime spectral selectivity also matters.
- Confirm OD blocking is sufficient for the ambient conditions. OD4 handles most outdoor sunlight rejection at NIR wavelengths; verify against the sensor, exposure settings, and measured ambient, and consider OD5 for very bright ambient.
- Check the filter's clear aperture against the sensor's active-area diagonal and the beam footprint at the filter plane so the filter does not vignette the lens's image circle over the sensor, and verify AOI-shift tolerance for short-focal-length or wide-angle lenses.
- Verify the lens is IR-corrected if switching between visible and NIR modes without refocusing is required.
- Validate with the real sensor, illuminator, and lens under representative ambient conditions. Specification review alone will miss AOI shift and daytime color-cast effects.
Häufig gestellte Fragen
What is a bandpass filter in machine vision?
A bandpass filter passes a defined wavelength band, such as 850nm plus or minus 25nm, and blocks wavelengths outside that band. Placed between the scene and the sensor, it isolates the illumination wavelength a system uses and rejects ambient light, other illuminators, and off-wavelength interference, improving contrast and signal-to-noise ratio.
What does an IR cut filter do in a machine vision camera?
An IR cut filter is a shortpass filter that blocks near-infrared light, typically above 650nm, before it reaches the image sensor. CMOS sensors remain significantly sensitive to NIR, and Bayer dyes pass NIR into all three color channels, so without blocking, NIR contaminates all three channels, causing washed-out color and incorrect white balance in daylight imaging.
What is a dual-bandpass filter in machine vision?
A dual-bandpass filter passes two spectral bands through one fixed element: the visible RGB range and a narrow NIR band such as 850nm, blocking the 700-800nm region between them. This lets a camera produce color images by day and NIR-illuminated images at night from the same static filter, with no mechanical switching component.
What is an IR cut filter switcher in machine vision?
An IR cut filter switcher is a small electromechanical assembly that moves an IR cut filter in and out of the optical path. In day mode the filter blocks NIR for accurate color; in night mode it is removed, replaced by a glass-matched dummy window, so the sensor regains full NIR sensitivity for illuminated night imaging.
What is the difference between a camera with an IR filter and one without?
A lens with an IR-cut filter installed (the 650/660nm IRC variant) blocks NIR for accurate daylight color. A lens without one (the NIR variant) passes NIR, which is correct only when the system deliberately uses 850nm or 940nm illumination. Neither fixed variant is a general-purpose upgrade over the other; the choice follows the illumination architecture.
What is the difference between an IR cut filter and a bandpass filter?
An IR cut filter blocks NIR above roughly 650nm to protect visible-color accuracy. A bandpass filter passes one narrow wavelength range and blocks everything else, isolating an active illumination wavelength such as 850nm from ambient light. They are not interchangeable: an IR cut filter has no role in NIR-illuminated imaging, and a bandpass filter cannot support daylight color imaging.
How do I choose a bandpass filter for machine vision?
Match the filter center wavelength (CWL) to the illuminator's peak emission from its datasheet. Choose a full-width at half-maximum (FWHM) wide enough to capture the LED emission and tolerate angle-of-incidence shift, but narrow enough to reject ambient. Confirm OD blocking is sufficient for the ambient conditions and that the filter's clear aperture covers the sensor's active-area diagonal and the beam footprint at the filter plane, so it does not vignette the lens's image circle over the sensor.
What do CWL and FWHM mean on a bandpass filter?
CWL is the center wavelength, the midpoint of the passband. FWHM is the full width at half maximum, the passband width measured at the 50% transmission points. A filter with CWL 850nm and FWHM 30nm passes roughly 835-865nm at greater than 50% transmission and attenuates wavelengths outside that range.
What does OD4 blocking mean on a filter?
OD stands for optical density. OD4 means the filter attenuates out-of-band light by a factor of 10,000, or less than 0.01% transmission outside the passband. OD4 is sufficient for most machine vision ambient rejection tasks, including outdoor sunlight rejection at NIR wavelengths; very bright ambient may call for OD5.
When should I use an 850nm vs 940nm bandpass filter?
Use 850nm when the illuminator emits at 850nm or when maximum sensor sensitivity matters, since silicon CMOS sensors typically have several times higher quantum efficiency at 850nm than at 940nm. Use 940nm when the illumination must be invisible to human observers, since 850nm LEDs emit a faint visible red glow and 940nm does not.
When should I use a mechanical IR cut switcher instead of a dual-bandpass filter?
Use a mechanical switcher when daytime color accuracy is a specification requirement. A switcher fully removes NIR from the path in day mode, while a passive dual-bandpass filter keeps its NIR passband open at all times, causing a measurable daytime color shift. Use the dual-bandpass filter when moving parts are not acceptable or the module has no room for a switcher assembly.
What does M12A650 mean compared with M12ANIR?
M12A650 and M12ANIR are Commonlands lens variant designators for the filter installed in the M12 filter slot. M12A650 has a 650nm IR-cut filter glued in place, for daylight color imaging. M12ANIR has no filter, passing the full spectral range the sensor detects, for NIR imaging, RGBIR sensors, and day/night operation with IR illumination. The lens optics are identical between variants; only the filter installation differs, and it is not field-removable. Confirm the current designation on the product page, since P/N conventions can vary slightly by lens family.
Can a bandpass filter reduce ambient light in machine vision?
Yes. A bandpass filter matched to the illumination wavelength rejects ambient light outside the passband, including broadband sources like sunlight, fluorescent, and incandescent lamps. The sensor receives mostly illumination light and little ambient contamination, improving contrast and stabilizing inspection results across changing ambient conditions.
Do bandpass filters reduce glare in machine vision?
Bandpass filters attenuate ambient light and off-wavelength reflections, but they do not eliminate specular glare from the system's own illumination source. If an 850nm LED reflects directly off a shiny surface into the lens, an 850nm bandpass filter passes that reflection rather than blocking it. Glare from the illumination itself requires cross-polarization, illumination geometry changes, or diffuse lighting.
Need help choosing a filter for your build?
Describe the illumination wavelength, sensor format, ambient conditions, and whether daytime color accuracy is a hard requirement. Commonlands engineering can help you pick between IR-cut, bandpass, dual-bandpass, and switcher architectures.