NIR camera shows a black image in low light: add IR illumination

You wired the camera, loaded the driver, opened the live view - and with the room lights off the frame goes almost black, with a faint shadow where the part should be. Nothing is broken. A near-infrared camera records only light that reaches it, and an ordinary room supplies almost none in the bands silicon sees at 850 nm and 940 nm. This page is for anyone bringing up a NIR or RGB-IR camera on a bench or in a cell. You will rank the likely causes, run five tests that isolate the fault, choose between 850 nm and the longer band, and size an illuminator to the field of view with a worked example.

Why the frame goes black

A NIR camera is not a passive night-vision device. The sensor turns photons into charge, and it can only turn photons that actually arrive. Indoors, almost none arrive in the near infrared: a white LED makes light with a blue pump and a phosphor, so its output collapses above roughly 700 nm, and fluorescent tubes behave much the same. Daylight and incandescent lamps are the exceptions - both are rich in NIR - which is why the same camera looks alive beside a window and dead in a windowless lab. The faint shadow you can barely see is filter leakage and read noise, not image signal. A camera that ships without emitters expects you to supply them.

Rank the causes before touching a setting

These five causes cover almost every bring-up, ordered by how often each turns out to be the fault:

1. No NIR source aimed at the scene. The kit or the cell simply has no emitter. Point a phone camera at the part with your light powered: no glow, no image.
2. NIR blocked before the sensor. An IR-cut filter sits in the lens or on the sensor stack, especially on hardware designed for colour. Remove it and brightness jumps by multiples.
3. Wrong pixels read. An RGB-IR sensor read as plain RGB buries the IR signal in pixels that were filtered against it. Pull the raw frame and check.
4. Exposure far too short, or the strobe fires outside it. Auto-exposure on a mostly black frame makes this worse. Set exposure and gain by hand.
5. Beam misses the field, or the band is mismatched. The light is on and aimed at the fixture, but its cone or wavelength does not match the camera. The coverage math below settles it.

Five bench tests, in order

Run these in sequence; each either finds the fault or removes a suspect. Test 1 needs nothing exotic: a 24 mm white spot on a 24 V supply covers it at bench range.

1. Light the scene with visible light. A normal image here proves the camera, lens, focus and exposure chain, leaving illumination as the only suspect.
2. Check your IR sources with a phone camera. Most phone sensors show 850 nm as a white-violet dot; at the band-edge wavelength they show little or nothing. No glow means no photons, whatever the power supply says - the phone-camera test for board illuminators walks through the details.
3. Set exposure and gain manually. Give the scene 20-50 ms and raise the gain until noise shows. Auto modes hunt badly on a black frame; if the first frames stay dark even with the light strobed, the settle-and-timing question is its own fault to chase, covered in the first-frame-dark guide.
4. Look for an IR-cut filter. Unmount the lens, check for a reddish-green glass in the path, and shoot with it removed.
5. Pull the raw frame on an RGB-IR part. If the clear-pixel channel lights up while the demosaiced image stays dark, your pipeline has been reading the filtered pixels all along.

A frame that stays black even in test 1 is not an illumination problem - jump to the last section.

850 nm or the invisible band: decide by the glow, budget for the band

Both bands work on silicon; the difference is the one the eye can see. 850 nm emitters leak a deep-red point of glow up close. Emitters at the longer wavelength show nothing. That single visible difference decides most purchases.

The 2x figure is a planning estimate with two causes stacking: silicon response falls toward the band edge, and an LED in the longer band radiates fewer milliwatts per amp than an 850 nm part of the same class. Budget it in exposure time or drive current. This page picks the source; the sensor-side half - exposure and gain under IR - pairs with the camera-module low-light guide.

Decision: pick 850 nm unless the glow is unacceptable. Where it is unacceptable - operator-facing benches, occupied rooms, discreet counting - pay the 2x and move on. If stray visible light pollutes the image once the IR source is in, a bandpass filter matched to the LED cleans it up; test the contrast before you buy, as the near-IR filter guide shows.

RGB-IR sensors: read the IR pixels

Many compact colour cameras use an RGB-IR colour filter array: a subset of the pixel sites - commonly around one in four - carry a clear filter tuned for NIR, while the rest carry R, G and B filters attenuated in the near infrared so colour stays accurate under mixed light. Under IR-only illumination the clear pixels collect nearly all the signal. A pipeline that demosaics the array as a plain RGB pattern therefore builds a dark, noisy image out of pixels that were filtered against the very light you supplied.

Fix it at the read, not at the light: grab the raw frame, or the IR channel your capture library exposes, and feed your tools from that.

Two expectations to set. First, the IR image from a shared array has lower spatial resolution than the RGB image would, because the clear pixels are a subset spread through the pattern, and that sparsity aliases fine detail. If the task needs full-resolution NIR, a monochrome NIR sensor is the cleaner tool. Second, never judge colour from an image taken under monochrome illumination. And if a colour image renders as plain monochrome, check which stream you are actually viewing - the black-and-white RGB guide on depth cameras walks through that cousin symptom.

Worked example: a 32 degree spot over a 160 mm field at 300 mm

The inputs, and where each comes from:

• Working distance 300 mm - the bench rig this example assumes; your lens fixes this number.
• Sensor width 6.4 mm - a 1/2 in optical format, that convention's active width.
• Focal length 12 mm - an assumed lens.
• Beam 32 degrees full angle - spot-light data sheets print lens angles, and 21 and 32 degrees both appear in the shop's spot range.

Two relations do all the work:

The 12 mm lens on the 6.4 mm sensor at 300 mm sees FOV = 300 mm x 6.4 mm / 12 mm = 160 mm. A 32 degree beam lays a disc of 2 x 300 mm x tan(16 deg) = 600 mm x 0.287 = 172 mm on the working plane. That clears the field with 172 - 160 = 12 mm of diameter margin, 6 mm per edge: it fits, but aim the beam square to the part.

The alternatives cost power. A 21 degree lens covers 2 x 300 mm x tan(10.5 deg) = 111 mm, which fails. Solving 2 x d x tan(10.5 deg) = 160 mm puts the light at d = 431 mm, where on-target power falls to (300 / 431)^2 = 0.48 of the 300 mm value - nearly half. Pulling the 32 degree head back to 450 mm buys 2 x 450 mm x tan(16 deg) = 258 mm of coverage for the same half-power penalty.

Decision: at a 300 mm working distance, take the wider beam. A spot clears this 160 mm field once its coverage passes about 170 mm; below that, widen the angle or move the light back and budget roughly twice the exposure to recover the lost power - and if you chose the invisible band, stack its 2x on top. If the geometry fights this hard, the field is too big for a spot; step up to a larger diffuse source.

The fix: order the source to the field

Three choices carry over from above: band, coverage and drive.

Band and size come straight from the lights finder: the spot category alone holds 44 models at 850 nm and 39 at the longer wavelength, filtered by wavelength and body size, and each data sheet prints the lens angle to check against your coverage number.

Where the field is small, raw irradiance wins. A 5 W high-luminance spot in a body under 50 mm puts everything the LEDs have into a tight cone, which is what a short exposure needs.

If the part moves, hold the light dark and fire it inside the exposure window. A dedicated strobe delivers far more peak intensity per pulse than a held-on lamp of the same average rating, and the pulse freezes motion; what an integrated driver changes is the strobe-driver guide's subject. The strobes and sequencing sources live with the special lights.

Drive the LED from current control, not a bare adjustable supply: output drifts with junction temperature under voltage drive, and the constant-current comparison shows by how much. A compact 5 V spot can take its pulse straight from the camera's strobe output within the port's current limit; the direct-drive wiring guide covers the basics.

For pulsed low-light capture, the timing sequence is its own subject and the pulsed IR timing guide owns it; this page stays with the source. To match light, camera and lens in one pass, start the configurator at the light step.

Keep the image bright on the next build

• Put the IR source in the bill of materials beside the camera. Bring-up then never starts dark, which is the whole failure mode this page fixes.
• Compute coverage before ordering. One division and one tangent, done in either direction by the focal-length calculator above.
• Lock exposure and gain in the recipe. Auto modes hunt on dark scenes and hide real changes; the auto-functions guide shows what they quietly alter.
• Strobe where you can. Pulses cut ambient, cut heat and freeze motion.
• Re-check alignment with a phone camera at maintenance. Beams drift when brackets get knocked, and the glow check takes ten seconds.

When the frame stays black under bright visible light

Then illumination is not your fault. Work the timing chain instead. A strobe that fires outside the exposure, or a trigger that never reaches the light, produces exactly this black frame; the strobe-timing guide orders the checks. A capture that comes back darker than the live preview is a source-settling problem in the software, and the darker-than-preview guide shows the fix. If the image is uniformly black rather than dark, check trigger polarity and the strobe output setting before you blame optics.

Once the visible-light test passes and the IR source is in place, the original symptom should be gone: a lit scene at the wavelength you chose, exposure locked, coverage confirmed. The lighting topic page collects the rest of the illumination patterns, and the application team can review a specific camera-and-light pairing through the contact page.