Book a demo
Global Shutter vs Rolling Shutter: Which Camera for a Moving Line

Global Shutter vs Rolling Shutter: Which Camera for a Moving Line

How each sensor treats a moving product, the blur and skew maths at belt speed, when a strobe rescues a rolling shutter, and a costed biscuit line example.
Global Shutter vs Rolling Shutter: Which Camera for a Moving Line

In short: A global shutter exposes every pixel of the sensor at the same moment and for the same time. A rolling shutter exposes the rows one after another, each for the same time, so a product that moves while the rows roll comes out leaning or stretched. That lean, called skew, is belt speed times the time the rows take to roll across the part. A shorter exposure cures motion blur on either sensor; it does nothing for skew.

  • A rolling shutter suits stations where the skew across the feature stays under a pixel at top speed, or where the part stops. A global shutter suits fast belts, belts whose speed changes, stations open to hall light and positions a robot will use.
  • A strobe rescues a rolling shutter only in a dark enclosure, and only in global reset mode or with an exposure longer than the rolling period.
  • Example plant: a rolling shutter corrected for one belt speed rejects 3.2 good biscuit packs per wrapper stop, $49,812 to $159,815 a year at UN trade and US retail values.

Guide, updated 16 November 2026. Sources are listed at the end.

The shutter is one decision at an inspection station, and not the first. Where the camera hangs, how far it sits from the product and how the light falls come before it; the guide to camera placement and lighting on a production line covers those. The shutter question comes up when the camera has to judge product that does not stop: packs on an outfeed belt, biscuits on an oven band, bottles on a conveyor. This guide works out what each kind of sensor does to a moving product, in pixels, then what a strobe can and cannot repair, then what each choice costs to run at one Example plant, priced with UN trade data and US retail prices. Everything below is about inspection cameras that pass or reject each product.

How each shutter exposes the rows

The GenICam feature standard defines the three modes in a line each. The GenICam Standard Features Naming Convention (SFNC) Version 2.7.1 from the European Machine Vision Association (EMVA), revision 2.7.1 dated 3 May 2022, gives its SensorShutterMode feature three values:

  • Global: the shutter opens and closes at the same time for all pixels, so every pixel is exposed for the same length of time, at the same time.
  • Rolling: the shutter opens and closes in sequence for groups of pixels, typically lines. Every pixel is exposed for the same length of time, but not at the same time.
  • GlobalReset: the shutter opens at the same time for all pixels but closes in sequence, so the pixels are exposed for different lengths of time.

SensorShutterMode is an optional feature, added in SFNC release 2.2 of 17 December 2014. A camera that implements it reports its shutter mode under that name, so a supplier can say in writing which of the three modes the camera offers.

Researchers at the University of California, Berkeley traced the rolling shutter to what a pixel can hold. In a 2005 paper on rolling shutter geometry they explained that the pixels of a CMOS chip typically could not store their charge, which forces each scanline to be exposed and read out in turn; as of 2004, they wrote, many CMOS chips in off the shelf cameras had no global shutter. Their point, made for robots that estimate motion from images, holds for any camera that measures: when the camera or the scene moves, a rolling shutter produces geometric distortion compared with a global shutter.

A 2026 paper from the Astronomical Institute of the Czech Academy of Sciences, written to correct meteor speeds measured with video cameras (Shrbený and others, arXiv preprint, published in Experimental Astronomy 62, 14), gives the timing terms in a form a plant can use. The time delay of a line is the interval between the starts of exposure of two consecutive rows. The rolling period is that delay times the number of rows, the time the sensor needs to start every row of a frame, and it can be much shorter than the time between frames. In their measurements the rolling period stayed the same when the exposure time changed. The same paper states the trade plainly: reading line by line makes higher resolution and higher frame rates possible, and it distorts fast moving objects, especially those moving with or against the direction of readout.

Three shutter modes, row by rowGreen: when each sensor row collects lightGlobalsame start, same lengthRollingsame length, each row laterGlobal resetsame start, rows end in turnAmber: a flash while every row is opentimeBlur= speed xexposureSkew= speed xrolling timeFlashonly whileall rows openShort exposure fixes blur.Only the shutter fixes skew.
How each SensorShutterMode exposes the rows, after the GenICam SFNC Version 2.7.1 definitions; bar lengths are illustrative. See the source.
Track OEE and downtime in real time
Book a demo

Two smears on a moving product: blur and skew

A moving product can be spoiled in two separate ways, and each needs its own fix.

Motion blur happens inside one row. While the row collects light, the product moves on, so a sharp edge is spread over several pixels. Both shutter types blur. The cure is a shorter exposure, set with the SFNC's ExposureTime feature, which controls how long the photosensitive cells are exposed to light, or a flash that is shorter than the exposure.

Blur (pixels) = belt speed (mm a second) x exposure (seconds) / pixel size on the product (mm)

Rolling shutter skew happens between rows. The top row of the part is captured at one moment and the bottom row later, once the rows have rolled down through the part. Only the rolling shutter skews. Exposure does not set the amount: the skew comes from the line delay, and in the Czech measurements the rolling period did not change with the exposure time.

Skew (mm) = belt speed (mm a second) x line delay (seconds) x rows the part covers

The direction of travel decides what skew looks like. A product moving along the rows, from side to side in the image, leans: its top and bottom are shifted against each other, and a square becomes a parallelogram. A product moving in the readout direction, up or down the image, is stretched or squashed instead. Rotation distorts too: the Berkeley paper's test images show a straight checkerboard warped when the camera turns about its own axis.

The pixel size on the product is the field of view divided by the pixel count along it: a 288 mm view on 1,920 pixels gives 0.15 mm a pixel. The line delay belongs to the camera and its settings. The Czech team writes that the rolling period is, in most cases, not part of the maker's camera specification, so ask for it in writing or measure it. They measured 15 settings of camera and resolution, on the video cameras of their own fireball network and two camera modules lent by another meteor network. The rolling periods ran from 14.01 ms to 38.66 ms, and the authors note the values are often around 16 or 32 ms. Those are video cameras, not industrial ones, so their numbers are no guide to a machine vision camera; the method is the lesson. One of their four methods used a short flash: the vertical shift of the edge between the lit and unlit parts of the image depends on the rolling period, though the authors judged their flash values not entirely correct.

The speed budget: pixels of smear at each belt speed

The table applies both formulas to one station geometry: 0.15 mm pixels, a feature 60 mm long across the belt, which covers 400 rows, and a line delay of 9 microseconds. The rows take 400 x 9 = 3,600 microseconds, or 3.6 ms, to roll across the feature. All of it is example data; put your own camera's line delay in its place.

Belt speedLongest exposure for 1 pixel of blurRolling shutter skew across the 60 mm featureGlobal shutter skew
9 m/min (150 mm a second)0.15 / 150 = 1,000 µs150 x 0.0036 = 0.54 mm, 3.6 pixels0
18 m/min (300 mm a second)500 µs1.08 mm, 7.2 pixels0
36 m/min (600 mm a second)250 µs2.16 mm, 14.4 pixels0
54 m/min (900 mm a second)166.7 µs3.24 mm, 21.6 pixels0
90 m/min (1,500 mm a second)100 µs5.4 mm, 36 pixels0

Example data: a 1,920 x 1,200 pixel sensor viewing 288 x 180 mm, with a 9 µs line delay.

The two columns call for different fixes. Blur is a light problem. EMVA 1288 Release 4.0 General, the EMVA standard for characterising image sensors and cameras (16 June 2021), defines the radiant exposure a sensor receives as the irradiance times the exposure time. For the same signal, halving the exposure means doubling the light on the sensor (our arithmetic). Skew is a sensor problem. At 54 m/min the lean across the feature is 3.24 mm whatever the exposure, more than three times a ±1 mm tolerance.

Our rule of thumb: when the skew across the feature at top speed is under one pixel, the shutter type does not decide the station. When it reaches the tolerance, as it does from 18 m/min up in this table, the choice is between a global shutter, a part that stops under the camera, and a rolling shutter with a strobe inside a dark hood.

When a strobe fixes a rolling shutter, and when it does not

A strobe is a light that fires a short, intense pulse timed to the camera. The SFNC describes the wiring: the sensor's exposure signal can be sent to an output line to control a strobe, and the camera's timers can generate a strobe pulse of a set length after a set delay. Its own example is a 200 microsecond pulse delayed by 100 microseconds. EMVA 1288 uses the same idea in its lab method with pulsed light: the camera's exposure is set longer than the LED pulse, and the light received is the LED irradiance times the pulse length. So when the flash is the only light that reaches the sensor, the flash length becomes the exposure, and that is how a flash freezes motion (our reading). Whether it also removes skew depends on the shutter mode.

Plain rolling mode: the flash lights a band

In rolling mode each row is open at a different time, so a short flash reaches only the rows that are open while it fires. With a 150 µs exposure, a 25 µs flash and a 9 µs line delay, about (150 + 25) / 9 = 19 rows see the flash: a band about 2.9 mm tall in a 180 mm view, while the rest of the frame sees only hall light. EMVA 1288 Release 4.0 General rules the pairing out for its own measurements, without giving a reason: its pulsed illumination method must not be used with rolling shutter mode.

Global reset mode: it works in the dark

In global reset, every row starts together, so a flash fired just after the start reaches every row with the same dose, and the pack is frozen in one 25 µs instant. EMVA 1288 asks for the same timing in its lab method: a short delay after the start of integration, with the pulse fitting inside the exposure. The catch is in the SFNC definition: the rows close one after another and are exposed for different lengths of time. Any light other than the flash keeps reaching the lower rows long after the flash is over.

On the Example plant camera, with a 40 µs exposure, the top row closes at 40 µs and the bottom row at 40 µs + 1,199 x 9 µs = 10.831 ms. For hall light to stay under 1% of the bottom row's signal, the flash at the sensor must be 99 x 10.831 / 0.025 = about 42,900 times brighter than the hall light. During those 10.831 ms the pack moves 900 x 0.010831 = 9.75 mm, so whatever hall light gets in paints a faint, skewed ghost of the pack about 65 pixels long. The arithmetic and the 1% threshold are ours, and they are why this guide pairs global reset with a hood that keeps hall light out.

Long exposure in rolling mode: dark again, and slower

A camera without global reset can still use a flash if each row's exposure is longer than the rolling period. There is then a window when the first row is still open and the last row has already started, and a flash in that window reaches every row. On the Example plant camera the exposure must be over 10.8 ms plus the flash. Every row then collects about 10.8 ms of hall light, so the same dark hood rule applies. The frame rate falls as well: the Czech paper's timing rule puts exposure plus row readout inside the time between frames, so with 10.8 ms of rolling period plus a 40 µs window for the 25 µs flash, 10.84 ms in all, the camera manages fewer than 1 / 0.01084 = 92 frames a second. That is enough for 3 packs a second, and not for a station that needs a frame every few milliseconds.

A global shutter with a flash: an open station can work

With a global shutter, every pixel opens and closes together, so hall light counts only during the exposure. With the same 40 µs exposure and 25 µs flash, the flash must be 99 x 0.04 / 0.025 = about 158 times brighter than the hall light for the same 1% threshold, against about 42,900 times for global reset. The ratio between the two, 10.831 / 0.04 = about 271, is what lets a global shutter station run behind a shield rather than inside a full hood (our reading).

When people work beside the light

A strobe that fires 3 times a second is a light that flickers 3 times a second. The UK Health and Safety Executive's guidance Lighting at work (HSG38, second edition, published 1997) calls light modulation at about 50 Hz or less, which is visible to most people, flicker. It says flicker can, depending on individual sensitivity, cause discomfort and fatigue, that it may even cause epileptic seizures in some people, and that it needs to be avoided. The guidance is written about workplace lighting, not vision strobes. On our reading, a strobe in an operator's line of sight still fits that description, so shield it from view, or choose a global shutter with a bright steady light and no flash at all.

The same guidance notes, under stroboscopic effects, that lamps on an alternating supply may produce oscillations in light output, though it calls the strong effect not common with modern lighting systems. Under a rolling shutter with a short exposure, rows exposed at different points of that oscillation can come out as brighter and darker bands (our reading). That is one more reason to keep hall light out of a rolling shutter station.

Decision table: which shutter for which station

Use one row per station. The cost column gives what each option spends or risks at the Example plant, where each good pack rejected or not made is valued at $0.72 to $2.31; the values are worked out in the example below. This guide found no independent price survey for camera hardware, so purchase prices are left to supplier quotes.

OptionGood atWhat it takes, and what it costs to runWho should pick itAsk the supplier
Rolling shutter, steady lightSlow belts and bands, high pixel countsA steady light and no enclosure. Nothing extra where skew stays under a pixel; at the Example plant outfeed, 3.2 good packs rejected per stop, $2.30 to $7.39 a stopStations where speed x line delay x rows across the feature stays under one pixel at top speedWhat is the line delay at the station's resolution, in writing? Does the skew correction read the belt speed for each image?
Any shutter, part stopped or indexedTray sealers, indexing tables, parts held in a fixtureA trigger from the machine when the part is still. No running cost from the camera; the machine's index time is the costMachines that already stop the part under the cameraWhich signal says the part has stopped, and for how long does it stay still?
Rolling shutter in global reset mode, strobe, hoodFast parts, with a camera that lists GlobalResetA strobe on the exposure or timer output and a hood that keeps the flash about 42,900 times brighter than hall light. Example plant: 20 s more per hood opening, 12 openings a shift, 720 packs not made, $518 to $1,663 a shift if the line is sold outEnclosed stations that operators do not need to open during a runDoes the camera list GlobalReset under SensorShutterMode? How is the flash delay set?
Rolling shutter, exposure longer than the rolling period, strobe, hoodCameras without global reset, a few parts a secondAn exposure above 10.8 ms plus the flash at the Example plant, the same dark hood, and at most about 92 frames a secondSlow part rates inside an enclosureWhat is the rolling period at the station's resolution? What frame rate at that exposure?
Global shutter, bright steady lightOpen stations with people nearby, belts that stop and startA steady light bright enough for 160 µs at 54 m/min, on whenever the line runs. No skew at any speed, so no ramp rejectsOutfeeds and conveyors next to operators, where speed changesEMVA 1288 data at the station's exposure: temporal dark noise, absolute sensitivity threshold, saturation capacity (quantum efficiency only from a Release 4.0 Linear sheet)
Global shutter with a strobeThe highest speeds, measurements and robot pick positionsA strobe on the exposure output, a shield from operators' view, a flash timing check after any changeBelts above 54 m/min, or any station where a robot uses the positionWhich output drives the strobe, with what pulse length and delay? EMVA 1288 data at the flash length

Then pick the line below that matches the station:

  • Choose a rolling shutter with a steady light if the skew across the feature at top speed is under one pixel, as on the Example plant's oven band.
  • Choose to stop the part if the machine already indexes it; then the shutter type hardly matters.
  • Choose global reset with a strobe if the camera lists the mode, the station can be closed against hall light, and operators do not open it during a run.
  • Choose a global shutter if the belt speed changes, the station is open to hall light or to people, or a robot or a measurement uses the position.
  • Choose a line scan camera instead if the product is a continuous web, such as film or foil; the area camera shutter is not the question there. How to size that station is covered in line scan vs area scan cameras.

A camera that only counts output is a different job with a different test: whether the count is right, not whether a seal is square. Fabrico's AI cameras (an add-on) count output at stations where a machine has no usable signal. They are not an inspection system. They do not check packs, read codes, recognise stops or propose reasons. How a count turns into availability and performance is covered in OEE with computer vision, and the counting tools are compared in the best computer vision tools for manufacturing OEE.

Worked example: an oven band and a flow wrapper outfeed

Example plant. Every figure in this section is example data except the two pack values, which are sourced. A biscuit plant bakes on a band oven and packs on a flow wrapper. Quality wants two camera stations: one on the oven band to check biscuit size and colour, and one on the wrapper outfeed to check that each pack's end seal is square to the pack within ±1 mm before the case packer. Both suppliers have offered the same 1,920 x 1,200 pixel rolling shutter camera with a 9 µs line delay, which gives a rolling period of 1,200 x 9 µs = 10.8 ms.

Station 1: the oven band at 7.2 m/min

The band runs at 120 mm a second. The camera sees 1,056 mm of band along the direction of travel and 660 mm across it, so a pixel is 1,056 / 1,920 = 0.55 mm on the biscuit, and a 62 mm biscuit covers about 113 rows.

  • Skew across one biscuit: 120 x 113 x 0.000009 = 0.12 mm, or 0.22 of a pixel.
  • Skew across the whole frame: 120 x 0.0108 = 1.296 mm, or 2.36 pixels, spread over the 660 mm across the band.
  • Blur at a 2 ms exposure: 120 x 0.002 = 0.24 mm, or 0.44 of a pixel. The longest exposure for one pixel of blur is 0.55 / 120 = 4.58 ms.

Neither smear shows against a size tolerance measured in millimetres. The rolling shutter camera with a steady light is right for this station, and a strobe or a hood would add cost for nothing.

Station 2: the wrapper outfeed at 54 m/min

The wrapper makes 180 packs of 200 g a minute on a 300 mm pitch, so the outfeed belt runs at 180 x 300 / 60 = 900 mm a second. The camera sees 288 mm along the belt and 180 mm across it, 0.15 mm a pixel, and the 60 mm end seal crosses 400 rows.

  • Skew along the seal: 900 x 400 x 0.000009 = 3.24 mm, or 21.6 pixels. A square seal looks crooked by more than three times the tolerance.
  • Blur: the longest exposure for one pixel is 0.15 / 900 = 166.7 µs.

The supplier's answer is a software correction that shifts each row back by the skew it expects. The Czech paper's correction needs the rolling period; a correction for a moving product needs the speed as well (our reading). This supplier's correction is set for 900 mm a second. At any other speed part of the skew is left over: (900 minus belt speed) x 0.0036 mm. The leftover passes 1 mm when the belt runs below 900 minus 1 / 0.0036 = 622 mm a second.

Skew left after a fixed speed correctionExample plant outfeed: 60 mm seal, 9 µs a rowSkew, mm01230300600900Belt speed, mm a secondRolling shutter, correctedfor 900 mm a second onlyOver 1 mm:good packs rejected1 mm seal toleranceGlobal shutter: no skew
Example data. Skew left over = (900 minus belt speed) x 0.0036 mm across the 400 rows of the seal.

The belt runs below 622 mm a second each time the wrapper stops and restarts. Example data: it stops in 1.5 s and restarts in 3 s, both at an even rate. It spends 1.5 x 622 / 900 = 1.037 s below 622 mm a second on the way down and 3 x 622 / 900 = 2.073 s on the way up. In that time it travels ½ x 622 x 1.037 = 323 mm and ½ x 622 x 2.073 = 645 mm, 968 mm in all, or 968 / 300 = 3.2 packs. Each of those packs shows a seal more than 1 mm out, and the camera rejects it.

The stop log shows 46 wrapper stops a shift. So:

  • False rejects a shift: 46 x 3.2 = 147.2 packs.
  • A year, at 2 shifts on 235 days, which is 470 shifts: 147.2 x 470 = 69,184 packs.

What a rejected pack is worth

  • Low: what the trade pays. UN Comtrade, the United Nations trade database, gives US imports of sweet biscuits in 2025 as $2,481,146,358 (CIF value) for 689,505,172 kg, a net weight that UN Comtrade does not mark as estimated. The heading is HS 1905.31, which the US Harmonized Tariff Schedule names "Sweet biscuits". That is $3.5984 a kg (our arithmetic), so a 200 g pack is 0.2 x 3.5984 = $0.7197, which the sums below use as $0.72.
  • High: what a shopper pays. The US Bureau of Labor Statistics average price for chocolate chip cookies, US city average, was $5.247 per lb (453.6 g) in October 2026. A 200 g pack is 200 / 453.6 = 0.44092 lb, so 0.44092 x 5.247 = $2.3135, which the sums below use as $2.31.

69,184 x 0.72 = $49,812 a year at the trade value, and 69,184 x 2.31 = $159,815 a year at the retail value, or 147.2 x 0.72 = $106 to 147.2 x 2.31 = $340 a shift. These are sales values, not the plant's cost or margin, and they assume rejected packs are scrapped. A plant that unwraps rejected packs and feeds the biscuits back loses less; this example does not price that rework.

The other options at the outfeed

Global reset with a strobe and a hood freezes the pack in 25 µs at any belt speed, so the ramp rejects go. The hood brings its own cost. Operators open it to thread film after a reel change and to clear jams at the outfeed; the stop log shows 12 such openings a shift, and closing and latching the hood adds 20 s to each, 4 minutes a shift. At 180 packs a minute that is 720 packs not made a shift: 720 x 0.72 = $518 to 720 x 2.31 = $1,663 if the wrapper's output is sold out, or 4 minutes of availability lost if it is not.

A global shutter with a bright steady light at a 160 µs exposure blurs a pack at full speed by 900 x 0.00016 = 0.144 mm, just under one pixel, and by less on every ramp. It has no skew at any speed, needs no hood and puts no flicker in front of the operators. Its cost is a light bright enough for 160 µs and whatever the camera costs over the rolling shutter one, which only quotes can show.

For this station the plant takes the global shutter with a steady light. The rolling shutter's ramp rejects would cost $49,812 to $159,815 a year, and the hood would add 4 minutes a shift to the stop chart. The choice would change only if the outfeed could be enclosed and nobody had to open it during a run.

On an OEE chart, the 147.2 packs a shift are quality loss, and their number moves with the count of stops: halve the wrapper's stops and the ramp rejects halve as well. That makes the stop count per shift a number to know before buying the camera, because it is also the number of ramps the station will see. OEE software can keep that count by line and shift from the machine signal.

Book a demo and see downtime per line and shift.

What each choice costs to run

The purchase price is one line of the cost. The table lists the others for the Example plant outfeed, valued where they can be at $0.72 to $2.31 a pack.

Cost to runRolling shutter, steady lightGlobal reset, strobe, hoodGlobal shutter, steady lightGlobal shutter, strobe
LightSteady light, on whenever the line runsStrobe, on for 0.0075% of the timeBrighter steady light, on whenever the line runsStrobe, on for 0.0075% of the time
EnclosureNoneLight tight hood, opened 12 times a shiftA glare shieldA shield from operators' view
Rejects caused by the shutter147.2 good packs a shift, on rampsNoneNoneNone
Time added to stopsNone4 minutes a shiftNoneNone
Checks to planLine delay and correction after any camera or speed changeHood seals and flash delayLight output, which can fall as the light agesFlash delay and output
Money a shift, Example plant$106 to $340$518 to $1,663 if sold outCamera and light quotesCamera and strobe quotes

Example data. The strobe is on for 0.000025 x 3 = 0.000075 of each second, 0.0075% of the time. In this example the money sits in rejects and minutes, not in the camera or the light, which is why the stop log is worth reading before the quotes.

The comparison starts from a fair case, not a weak one: a supplier who already corrects skew in software and a crew that opens the hood only when it must. Set against a station with no check at all, any camera would look good and the sums would tell a buyer nothing.

What to ask before you sign

A supplier who knows the station can answer each of these in writing. Send them with the speed budget so that every quote answers the same questions.

  1. The line delay or rolling period at the station's resolution. If it is not on the datasheet, ask how it was measured. The Czech team compared four bench methods, from a flashing light to timed blinking diodes, and found the diodes synchronised to satellite time the most accurate.
  2. The shutter modes the camera offers, by their SFNC names: Global, Rolling or GlobalReset.
  3. How exposure is controlled. The SFNC's ExposureMode values include Timed, where ExposureTime sets the duration, and TriggerWidth, where the width of the trigger pulse sets it. Ask which one the station will use and who sets it.
  4. How the strobe is driven: from the exposure signal on an output line, or from a timer with a set delay and pulse length, and what those values are.
  5. EMVA 1288 data at the station's exposure for every camera offered. The Release 4.0 General summary sheet reports temporal dark noise, the absolute sensitivity threshold and the saturation capacity for each operating point with its exposure time; it lists quantum efficiency as N/A, because the general model cannot measure it, while Release 4.0 Linear includes it. This guide makes no general claim about which shutter type is more sensitive; compare the sheets.
  6. How any skew correction gets the belt speed: a fixed value, or a reading for each image from an encoder or the machine. Ask what happens on ramps.
  7. What the station does while a hood is open, and what happens to the packs made then.

Then write three tests at line speed into the order. The pass marks are our advice.

  1. Ramp test. Stop and restart the line ten times with known good packs. Pass: no rejects caused by the ramps.
  2. Square target test. Run a pack printed with a square at rated speed. Pass: the measured corners are square within a tenth of the tolerance.
  3. Open station test. Run with the hood open, or with the hall lights at full, and compare with the closed or dark result. Pass: no ghost, no band and no change in the decision.

Before the purchase order: steps for plant, quality and CI managers

This week

  1. Plant or production manager: for each planned camera station, write down the top belt speed and whether the product stops under the camera.
  2. Quality manager: write the tolerance for the feature the camera must judge, and how long that feature is across the belt in millimetres.
  3. CI manager: from the stop log, count the stops per shift on each station's machine. That count is the number of ramps the camera will see.

This quarter

  1. Build the speed budget for each station: pixel size on the product, line delay, skew across the feature and longest exposure at top speed. Send it with the request.
  2. Ask every supplier the seven questions above, and treat a rolling shutter quote without a line delay as incomplete (our advice).
  3. Put the three tests in the purchase order, with the belt speeds they run at.
  4. Decide where the check belongs. A check can also sit at the end of the line or in the lab instead of on a fast belt; inline vs offline inspection sets out the trade.
  5. Cut the stops that cause the ramps. Log wrapper stops with a time and a reason, and track the rejects next to them by line and shift, so the cost of a stop includes the packs the camera throws out after it.

Where Fabrico sits beside the inspection station

Fabrico is a cloud OEE and light MES platform: OEE and losses by machine, line and shift, from the PLC or a Fabrico IoT device, with the reasons operators add.

On manual stations, AI cameras (an add-on) count output.

Book a demo and see OEE and losses per line and shift.

What people ask next

What is the difference between a global shutter and a rolling shutter?

A global shutter opens and closes all pixels at the same moment, so the whole image is one instant. A rolling shutter exposes the rows one after another, each for the same length of time, so a moving product is captured at slightly different moments from top to bottom and can lean or stretch. These are the definitions in the GenICam SFNC, the EMVA's camera feature standard.

Is a rolling shutter camera good enough for a conveyor?

It can be. Multiply the belt speed by the line delay and by the rows the feature covers. If the result is under one pixel at top speed, as on the Example plant's oven band at 7.2 m/min, a rolling shutter is fine. At the Example plant's outfeed, 54 m/min, the same camera leans a 60 mm seal by 3.24 mm, over three times a 1 mm tolerance, and a global shutter suits the station.

Can a strobe light fix rolling shutter distortion?

Only in two set-ups, and both need the dark: global reset mode, where all rows start together and the flash fires while every row is open, or an exposure longer than the rolling period with the flash in the window when every row is open. In plain rolling mode a short flash lights only a band of rows. Hall light keeps reaching the rows for longer than the flash, so the station needs a hood.

How do I work out motion blur on a moving line?

Blur in pixels is the belt speed times the exposure time, divided by the pixel size on the product. At 900 mm a second with 0.15 mm pixels, an exposure of 166.7 microseconds gives one pixel of blur. A shorter exposure or a flash reduces blur on either shutter type; it does not reduce rolling shutter skew.

What is global reset mode on a camera?

A shutter mode, named GlobalReset in the GenICam SFNC, in which every row starts its exposure at the same moment and the rows end one after another, so they are exposed for different lengths of time. With a flash fired while every row is open, inside a dark enclosure, it gives a frozen image without skew.

What should I ask a camera supplier about the shutter?

The line delay or rolling period at your resolution, in writing; which shutter modes the camera offers; how the strobe output is driven; EMVA 1288 data at your exposure; and whether any skew correction reads the belt speed for each image. A 2026 paper from the Czech Academy of Sciences, measuring video cameras, notes that the rolling period is, in most cases, not part of the maker's camera specification.

Sources

Sources accessed on 16 November 2026 unless marked as accessed on another date. Dates are publication dates where a page gives one; other dates are labelled as what they are.

Track OEE and downtime in real time
See where lines lose time
Book a live demo.
See where lines lose time
See where lines lose time

Real Fabrico tiles, sample data.

By clicking the Accept button, you are giving your consent to the use of cookies when accessing this website and utilizing our services. To learn more about how cookies are used and managed, please refer to our Privacy Policy and Cookies Declaration