IR LED eye safety at close range: IEC 62471 limits

A 940 nm IR flood pointed straight at a face — eye tracking at 0.4 m, face recognition, a depth assist — raises the question that stops installations: is it eye-safe? The question has a numerical answer. IEC 62471:2006, the photobiological safety standard for lamps and illuminators, ties every hazard value to a declared measurement distance, so the check is arithmetic: read the W/m² value the report states at its declared distance, scale it to the closest eye position with the inverse-square law, and compare against a stated limit. This guide walks that check for a face at that working distance, covers the reference-plane detail that sets the declared distance, and ends with the mounting distance each case allows. For the depth-assist variant, see infrared flood illumination for stereo depth imaging; for illumination stability, fixing illumination drift in face-recognition lighting.

Which number decides eye safety for a close-range IR illuminator?

Irradiance in W/m² at a stated distance decides it: IEC 62471:2006 evaluates each non-general-lighting source at 200 mm from its reference plane, and a face held at 0.4 m receives a quarter of the value stated there.

Watts alone decide nothing. The same 8 W of LED output spread across an 80° × 80° flood cone and focused into a 10° spot expose an eye at the design distance to irradiances orders of magnitude apart. That is why the standard ignores raw power and measures irradiance in W/m² at a fixed point, plus radiance in W/(m²·sr) for the retinal metric, and every figure it reports belongs to a distance. LED data sheets rarely print that distance, which is how watt ratings end up compared against limits they were never meant to meet. The measured hazard values collapse into four risk groups, and those groups are what you actually order against.

What do IEC 62471 risk groups mean for face lighting?

Only Exempt (RG0) and Risk Group 1 are acceptable when an invisible beam faces a person: RG2 depends on the blink and look-away reflexes that an invisible source at this wavelength can never trigger, and RG3 must never be aimed at a person.

A group is declared at the measurement distance, not for all distances: step inside it and a comfortable RG1 source can cross its limit line, which is why the scaling in the next sections matters. For face lighting the practical shortlist is Exempt or RG1 at the closest eye position, with the geometry held by the mount rather than by discipline.

Which hazard governs at 940 nm?

At 940 nm the governing hazard is heating of the cornea and lens, an irradiance quantity that falls with the square of distance, so extra working distance buys real margin.

The wavelength sits in the 780–1400 nm band the standard assesses as an infrared hazard to the eye. For continuous exposure, the occupational guidance behind most illuminator safety notes — the ACGIH threshold limit values, aligned with ICNIRP — caps corneal irradiance in that band at 100 W/m², or 10 mW/cm². Use it as the design line: brief glances tolerate more, but a set-up that stays under the line at the closest eye position removes the argument entirely. Two refinements keep the number honest. The retinal metric is radiance, which for an extended source does not fall as the eye moves back, but the retinal weighting drops steeply toward the long-wave end, so the corneal irradiance comparison governs a wide flood. And design with margin: 20 W/m² or less at the closest eye position, a five-times cushion.

How do you scale the report value to your working distance?

Scale by the inverse square of the distance ratio: a hazard value E_d quoted at the declared distance d becomes E_d × (d/r)² at distance r, so moving a face from 0.2 m to 0.4 m cuts the irradiance to a quarter.

Worked example, with a flood whose report declares Risk Group 1 and E = 40 W/m² at the standard 200 mm (example figures, typical of a duty-cycled wide-angle unit):

• Face at the design distance: E = 40 × (200/400)² = 40 × 0.25 = 10 W/m², one tenth of the line. Decision: continuous face illumination is acceptable with a ten-times margin.
• Face leaning in to 0.1 m from the LED: E = 40 × (200/100)² = 40 × 4 = 160 W/m², over the line. Decision: unacceptable without a time limit, so the mount must prevent this geometry.
• Break-even distances: the exposure line sits at r = 200 mm × √(40/100) = 126 mm from the LED, and the 20 W/m² design target at r = 200 mm × √(40/20) = 283 mm.

The second route covers the case with no report yet. Estimate on-axis irradiance from the data sheet's radiant intensity: with I = 0.8 W/sr, a face at 0.4 m receives E ≈ 0.8 / 0.16 = 5 W/m². Near an extended panel this over-predicts, because far-field radiant intensity divided by r² is an upper bound once r approaches the panel's own size, so the error sits on the safe side.

Why does a report declare 194 mm instead of 200 mm?

Because 200 mm counts from the source's reference plane, not from the housing: a flood LED sitting 5.7 mm behind the cover window is classified at 194 mm from the window, and that declared distance is the one your scaling starts from.

The limit tables print W/m² and W/(m²·sr) with no distance attached because the distance lives in the measurement conditions: 200 mm from the reference plane for every non-general-lighting source, which is why a skim of the tables finds no distance. For general-lighting lamps, EN 62471 instead moves the measurement out to where the lamp produces 500 lx when that is farther; an IR illuminator is not general lighting, so the fixed distance applies. Read the declared distance off every report and scale from that number, not from the housing face: with the 194 mm declaration above, the correction is (200/194)² = 1.06, about six percent, small but real, and larger whenever optics bury the die deeper. A report citing IEC 62778 is applying that standard's method to the visible-light portion; the infrared hazard values still come from 62471.

Is skin heating a concern at face range?

No: the 10 W/m² the example delivers at the face is about one percent of summer sunshine, roughly 1000 W/m² at ground level, so skin warming is imperceptible and the eye stays the governing organ.

The standard's skin-thermal limits sit orders of magnitude above these irradiance levels, and at the working range the face would feel warmth long before a hazard threshold came near. When you want evidence for the installation file, image the face during a full-power test: a thermal camera such as the LS-IRC-0044 records 640 × 512 pixels across the 8–14 µm band at 50 Hz over GigE, enough to show whether the flood adds anything measurable to skin temperature. Converting those counts to degrees needs the calibration route in raw thermal counts to temperature: calibration essentials; the infrared cameras category lists the range and the camera finder filters it.

Does pulsed operation reduce the hazard?

Yes, in proportion to duty cycle: a flood driven 1 ms on and 32 ms off delivers about 3 percent of its peak as an average, and the guideline compares against the average over minutes, not the peak.

Average irradiance is what heats cornea, lens and skin, and duty cycle divides it: 1/33 ≈ 0.03, so a peak of 300 W/m² at the eye works out to 9 W/m² average. Two cautions keep this honest. Peak pulse irradiance still drives instantaneous retinal heating, so do not push drive current hard to buy the average down; stable drive design is its own subject, covered in LED overdrive and current control and in the wider machine vision lighting operating modes comparison. Pulse timing also has to land inside the camera's exposure, which pulsed IR illumination timing for low-light capture works through in detail. For the driver-side choices of modulation frequency, depth and range, this IR LED modulation and range design reference goes deeper.

How do you verify the set-up before a face goes in front?

Order the photobiological report with the light, confirm its risk group and declared distance, then hold the closest eye position at the distance your scaling gives for a five-times margin: 20 W/m² or less against the guideline.

1. Obtain the photobiological test report with the light: risk group, hazard values, declared measurement distance, reference plane.
2. Scale the worst hazard value to the closest eye position the mount allows, and require 20 W/m² or less against the 100 W/m² line.
3. Fix the geometry so it cannot drift: a bracketed mount, a mechanical stop, or a housing that sets the face distance.
4. View the beam with a camera, never with your eye: SWIR sensors image the wavelength directly, and a silicon phone camera picks up a trace of it, as the phone-camera illuminator test demonstrates.
5. If someone must work at arm's length during alignment, issue eyewear with a stated optical density at 940 nm; the background on choosing LED safety glasses helps there.

For the beam view an SWIR camera is the direct instrument, and the two below cover different resolution needs.

All 2 share: Lens mount C-mount. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

The decisions, one line each: a report-backed RG1 flood at a face 0.4 m away is acceptable continuously; the same flood fails inside roughly 130 mm, so the mount needs a stop at 300 mm or more; duty cycling buys margin back in proportion. To pick hardware, start at the special lights category or the lighting finder, or go straight to the lighting step of the configurator; we quote custom floods built to a declared irradiance when a standard housing does not fit. More illumination guides sit under the lighting topic.