How strobe pulse width freezes motion blur in machine vision

During a strobed capture, the light pulse — not the camera's exposure setting — is the effective exposure time. That single fact turns motion blur into arithmetic: blur equals part speed times pulse width. A part crossing the lens at 300 mm/s keeps its blur inside a quarter of a 0.02 mm pixel only when the pulse stays near 17 µs, while a 2 ms continuous light smears the same part across 30 pixels. This guide walks the pulse budget from line speed and pixel size, shows the brightness arithmetic that pays for a short pulse, flags the rolling-shutter trap, and closes with the integrated-driver versus external-controller decision and the checks to finish before you commit to parts.

Why the pulse sets the blur, not the shutter

The camera may hold its shutter open for 2 ms. The light fires only for the budgeted pulse inside that window. Nearly every photon that forms the image arrives during those microseconds, so nearly all the smear happens then too. The exposure setting controls how much ambient light and sensor noise you collect; the pulse width controls how sharp the part is.

Two timing rules make this work. The pulse must fall entirely inside the exposure window, and the exposure should not run far past it, or ambient light keeps smearing the image after the pulse has ended. Cameras that let the trigger pulse itself act as the exposure merge the two numbers into one wire; the mechanics are in why trigger pulse width sets your camera exposure time, and the moment-by-moment sequence in trigger call versus exposure time. When a triggered frame still comes out black, the pulse usually missed the window: first frame dark: settling or timing separates the two causes.

On a global-shutter sensor the arithmetic then holds for the whole frame:

Here v is part speed in mm/s, t_pulse the light-on time in seconds, p the object-space pixel size in mm, and k the blur you accept in pixels. Presence checks, code reading and OCR tolerate k = 1 pixel. Gauging, alignment and robot guidance want k = 0.25. The object-space pixel size calculation gives p from magnification in one line if you have not computed it yet.

The pulse budget at a glance

Pixel size drives everything, so fix it first. The table below assumes a 50 mm field on a 2448-pixel-wide 5 MP sensor, which works out to p = 50 mm ÷ 2448 ≈ 0.02 mm per pixel; double your pixel size and every number in it doubles.

Read the row that matches your belt and stay under it. At 300 mm/s a presence loop is comfortable at 60 µs; a gauging station needs the 17 µs from the table or better. Below 10 µs, subtract the light's rise time from the budget — a one-microsecond rise on a 5 µs pulse is a fifth of your light-on time spent getting bright. Why microseconds of rise time smear short pulses shows how to measure it on your own light. The rest of the illumination arithmetic sits on the lighting topic page.

Worked example: a 300 mm/s conveyor

Setup: parts on a belt at 300 mm/s; a 5 MP camera, 2448 pixels wide, covers a 50 mm field; the job is a dimension check, so k = 0.25.

1. Pixel size: p = 50 mm ÷ 2448 = 0.0204 mm, rounded to 0.02 mm.
2. Blur budget: 0.25 × 0.02 mm = 0.005 mm.
3. Pulse ceiling: t = 0.005 mm ÷ 300 mm/s = 16.7 µs. Order a 16 µs pulse; controllers step in whole microseconds.
4. Check the result: blur = 300 mm/s × 16 µs = 0.0048 mm = 0.24 pixels, just inside the 0.25 ceiling.
5. Compare continuous light: the test frame needed 2 ms, so blur = 300 mm/s × 2 ms = 0.6 mm = 30 pixels. The strobe is 2 ms ÷ 16 µs = 125× shorter, which is the whole point of putting it on this line.

The figure has margin: a 20 µs pulse gives 0.006 mm, still only 0.3 pixels, so a few microseconds of trigger jitter will not sink the image. Duty cycle is a non-issue at conveyor rates — the 17 µs ceiling × 1,000 triggers/s is only 1.7% duty, and the full duty-cycle matching math shows where it starts to bind.

Paying for the short pulse: brightness, aperture, gain

A 125× shorter exposure gathers 125× fewer photons unless the light gets brighter in the same ratio. Overdrive strobing is that trade: pulse the LEDs at several times their continuous current — typically 5 to 10 times the continuous brightness — and let the driver's current limit and duty-cycle protection keep the emitters alive. The practice, its limits and its timing are laid out in strobe overdrive, duty cycle and timing.

Run the conveyor numbers. Continuous, the scene wanted 2 ms. Driven 10× harder at that pulse width, the pulse delivers the light of a 160 µs continuous exposure — 8% of what the scene asked for, a shortfall of 12.5×. Two camera-side dials close it:

• Aperture: f/5.6 to f/2.8 is two stops, 4× the light. The cost is depth of field, which shrinks by the same two stops; run the new f-number through the depth of field calculator before committing.
• Gain: 10 dB is about 3.2× brighter pixel values. The cost is mild extra noise — usually an easier trade than adding hardware.

4 × 3.2 = 12.8 ≥ 12.5, so the frame lands at normal brightness. If the gap were wider than two stops plus 10 dB, add hardware instead: a closer head, a wider reflector or a second light beats fighting sensor noise.

Ambient light: the second job of the pulse

Strobe also cleans up images in a bright plant, and how much it helps depends on one wiring choice:

• Exposure equals the pulse (trigger-width exposure): ambient light integrates for the same few microseconds as the strobe. This is the strongest ambient rejection per microsecond of light and the simplest to reason about; if the camera still runs auto exposure, cap the maximum auto exposure so the window cannot creep open.
• Exposure stays open (2 ms) while the light fires the pulse: ambient now integrates 125× longer than the signal, so the strobe must outshine the plant lighting by roughly 125× at the sensor, or a smeared ambient ghost overlays your sharp image. Some stations expose twice on purpose — one strobed, one ambient-lit — and the ambient light rejection specification is how acceptance tests pin that down.

On a fast line, prefer the first configuration: the strobe then acts as shutter and ambient filter at once.

Global shutter or rolling shutter

A global shutter samples every row at the same instant, so one pulse freezes the whole frame and the blur budget applies everywhere. A rolling shutter exposes row by row. To light a full frame you must hold the pulse across the whole scan — on the order of 30 ms for a 30 fps sensor — and 300 mm/s × 30 ms = 9 mm of blur. The strobe advantage is gone.

So the rule is simple: anything that moves faster than a walk gets a global shutter. The MV-CS200-10UM is a 20 MP USB3 camera with a rolling shutter — excellent on static parts, the wrong shutter for this belt. Two global-shutter cameras from the area scan cameras range, one compact GigE and one 12 MP high-speed USB3:

If a rolling-shutter camera must stay, shrink the region of interest: halving the rows roughly halves the scan time and the blur that rides on it. Browse more global-shutter options in the camera finder, and if the words exposure and shutter speed are tangled, exposure time versus shutter speed untangles the naming.

Integrated driver or external controller

Two architectures produce the pulse:

• Integrated driver: the light carries its own strobe electronics, and a trigger line from the PLC — or straight from a camera GPIO — fires it. One box fewer, ideal for a single light of modest power. If the trigger needs a breakout, do I need a camera GPIO cable settles it, and camera trigger lines as digital outputs frees a line when the PLC has none spare.
• External controller: a box between PLC and lights supplies per-channel power, sequencing and one control interface for many heads. Take this route when several lights fire in sequence, when the light's own driver cannot deliver the current your pulse budget needs, or when one station runs many heads.

The controller's role in blur is narrow: hold the requested width repeatably, down to a few microseconds. Controllers of this class hold 10-20 µs comfortably; below about 2 µs the LED's rise time, not the controller, sets the effective width. Width applies per channel, so a sequenced station keeps the same pulse-width rule on every head.

Controllers that cover the three cases

For a general-purpose strobe station with digital control:

The LS-CTL-0070 is a 4-channel, 200 W digital controller with a strobe feature and RS232 control, delivering constant voltage 24 V.

For strobed bars and rings, constant-current drive holds LED brightness flat through the pulse:

The LS-CTL-0180 is a 4-channel constant-current controller rated 120 W, 89 × 147 mm, for 24 V lights.

And for large diffuse heads driven hard in short bursts:

The LS-CTL-0054 is a 2-channel high-power constant-current controller rated 240 W per channel, 320 W overall.

The full range sits on the light controllers category page, filterable in the light controller finder.

When strobing is not the answer

• Line scan: a line camera exposes each line for one line period, often tens of microseconds, so motion is already frozen per line and the fight is photon supply, not blur — see how much power a fast line really needs and lighting controllers for line scan cameras.
• Dark enclosure: continuous light plus a capped short exposure freezes motion identically, with no overdrive arithmetic. The strobe earns its place when the camera cannot expose shorter than a few milliseconds — no software exposure control: what to do instead — or when ambient light demands it.
• Fast rotation: the blur math uses surface speed. A 50 mm roller at 3,000 rpm moves its surface at π × 50 mm × 50 rev/s ≈ 7,850 mm/s, and the 0.25-pixel budget collapses to 0.005 mm ÷ 7,850 mm/s ≈ 0.6 µs. There you are shopping for a faster light or switching to a continuous head with a short electronic shutter, not comparing controllers.

Everything else on a moving line — presence, codes, gauging, robot guidance — is strobe territory, and the budget above sizes it.

Before you order: the five checks

• Compute pixel size from field of view and sensor pixels. Without p the budget is a guess.
• Fix k by task — 1 pixel for presence and codes, 0.25 for measurement — and divide by line speed for the pulse ceiling.
• Read the light's strobe data: minimum pulse width, rise time, and the duty cycle allowed at your pulse current. Rise time is the hidden tax under 10 µs.
• Confirm the shutter: global for anything that moves; if rolling, budget the full scan or shrink the ROI.
• Size the controller: at least as many channels as heads, wattage above the connected load, strobe feature present, and an interface your PLC speaks.

Send line speed, field of view and trigger rate through the controller configurator, or bring the numbers to the contact page and the team will match a light and controller to the budget before you commit to parts.