When the Light Lies: Flicker, Drivers and the Frames Your ANPR Camera Never Sees

Stand on a UK forecourt at three in the morning and the lighting looks immaculate: even, bright, steady. Point an ANPR camera at the same surface with a 1/500 s exposure and the frame may come back with dark bars sliding across the plate, half the characters washed out and the recognition engine declining to commit. Nothing in the camera is broken. Nothing in the luminaire looks wrong. The fault sits in the relationship between the two, and it is one that procurement rarely tests because lighting and surveillance are almost always bought as separate packages.

Two Devices Sharing One Waveform

An LED luminaire is not a constant light source. It is a semiconductor junction fed by a driver, and that driver shape-shifts the output continuously. There are two mechanisms worth separating.

The first is pulse-width modulation. To dim an LED, many drivers switch the output on and off rapidly and vary the proportion of on-time. At a few hundred hertz this is invisible to human vision, which integrates over roughly a fiftieth of a second. It is emphatically not invisible to a camera with a short exposure, because a short exposure samples a narrow slice of a repeating waveform and can land entirely in the off portion.

The second is ripple. The driver rectifies 50 Hz mains into direct current, and unless that DC is well smoothed it carries a residual oscillation at 100 Hz — twice mains frequency, because both half-cycles of the AC supply contribute. Pools of light on a site fed by mediocre drivers therefore pulse at 100 Hz whether or not anyone has dimmed anything. This matters for more than installation quality: electrolytic capacitors age, and a driver that showed 5 per cent ripple at commissioning with a comfortable temperature profile can present 20 per cent or more after several summers in a roadside cabinet.

What the Camera Does With That

Most modern IP cameras use rolling-shutter CMOS sensors. The sensor does not capture the whole frame at once; it reads lines sequentially over a readout period that can be comparable to the exposure itself. If the illumination is varying at a frequency that is not harmonised with the readout, adjacent lines receive different amounts of light and the frame acquires bands.

Two camera settings then collide. Anti-flicker mode, where it exists, locks exposure to a multiple of the mains period — typically 1/50 s or 1/100 s — which suppresses 100 Hz artefacts neatly and renders the image stable. But plate recognition needs short exposures to freeze a vehicle moving at 20 or 30 mph through the capture zone, so the anti-flicker setting is usually switched off or overridden. WDR compounds the problem by combining multiple exposures of differing length, each sampling the waveform differently. Automatic gain then amplifies whatever inconsistency survives.

The consequences extend past a single unreadable plate. Colour rendition drifts frame to frame, white balance hunts, analytics confidence scores fall, and human reviewers in a control room begin to distrust footage that is technically intact but practically marginal.

The Figures Worth Writing Into the Specification

Flicker is not a matter of opinion; it has defined metrics. Percent flicker, or modulation depth, expresses peak-to-trough variation as a proportion of the total. Flicker index compares the area above the mean to the total area under the waveform. Both are measurable and both should be declared, at every dimming level the site will actually use — not only at full output, where drivers invariably perform best.

For areas covered by recognition or identification cameras, a sensible procurement position is:

• Modulation depth no greater than 3 per cent at 100, 50 and 10 per cent output, with 1 per cent or better in the plate capture zone.
• Driver output ripple current no greater than 3 per cent, evidenced on a datasheet rather than asserted in correspondence.
• Where PWM is used for dimming, a switching frequency of 4 kHz or above; amplitude or constant-current reduction dimming is preferable where offered.
• Compliance with the retained ecodesign requirements for light sources on temporal light modulation, and a declared statement of behaviour under emergency inverter operation, where the luminaire forms part of a life-safety circuit.

That last point is often overlooked. An emergency conversion module running from a battery inverter is a different supply with a different frequency, and a luminaire that behaves impeccably on mains can strobe on battery.

Eight Questions for the Lighting Supplier

Ask these before the order is placed, and ask for answers in writing:

1. Is dimming achieved by PWM or by current reduction?
2. If PWM, at what frequency, and does that frequency change with dim level?
3. What is the output ripple current at 100 per cent, 50 per cent and 10 per cent load?
4. What is the declared modulation depth at those same levels?
5. How does ripple degrade with capacitor ageing over the stated driver life?
6. What is the driver behaviour below 10 per cent output, where many sites run overnight?
7. How does the luminaire behave on an emergency inverter or a central management system dimming command?
8. Can the manufacturer supply a photodiode trace or test report rather than a marketing claim?

A supplier who cannot answer questions two and three is not necessarily selling poor product, but the buyer has no basis for confidence and should price the risk accordingly.

A Banding Test You Can Run Before Handover

The test is inexpensive and takes an evening. It should be written into the commissioning programme, not left to goodwill.

Mount the actual camera model, on its final firmware and with the settings intended for service, on a tripod aimed at the lit surface. Set the exposure to the value the site will use — commonly 1/500 s for a plate zone — and disable anti-flicker. Drive the luminaire through 100, 50, 20 and 10 per cent output, recording at least 60 seconds at each level. Repeat with WDR enabled, and repeat at the coldest and warmest ambient temperatures the site will plausibly see.

Review single extracted frames, not the smooth video stream, since the eye will average away precisely the artefact being hunted. A photodiode and oscilloscope will give a waveform for the record, but frame review alone will settle most disputes. Pass criteria should be stated numerically in advance: no more than a defined luminance variation across the capture zone at any dim level, and recognition of a test plate at the specified approach speed.

Repeat the exercise at the first annual service. Capacitor ageing is gradual and site conditions vary; a luminaire estate that passed in December may not pass in its third winter.

One Clause, Not Two

The underlying lesson is contractual. When lighting and surveillance are specified separately, each package can be fully compliant while the combination fails. The remedy is a single acceptance clause spanning both, requiring a joint mock-up, declaring the camera settings the luminaire must support, and holding the lighting supplier to a flicker performance figure for the life of the driver warranty. On a UK site where plates, faces and evidence matter, that clause is worth more than another megapixel.