Deciding whether a rolling shutter is good enough for your moving part

In short

Motion alone does not disqualify a rolling shutter. Name the dimension you measure and its reference points, allot the shutter a share of its tolerance, and get the image-formation timing of their rows in the mode you will run. Displace each point by the part's travel over that difference, recompute the dimension, and compare the change with the share you allotted. Budget the blur inside one row's exposure separately: a global shutter removes the row-to-row term and leaves blur standing. A rolling shutter passes under stable continuous light when both stay inside budget, and under a shared pulse when every required row is sensitive together; confirm either on a calibrated rigid target at worst-case speed.

The difference is when each row forms its image

A global shutter exposes all rows over one common interval, while a rolling shutter starts them in turn, which Sony describes as line-sequential capture against capture of the whole focal plane in one shot. All three cameras below are CMOS (datasheet), so the shutter field settles this, not the technology name.

The distortion attributed to a rolling shutter comes from that offset in time, not from exposure smear. Basler states that fast motion distorts the image because rows start at different times, and exposes a model-specific parameter, SensorReadoutTime, for the readout span of the rows you take; it belongs to the model and its configuration, not to the exposure setting or the frame rate. Allied Vision adds that rows in a conventional rolling mode can share one exposure duration yet begin at different moments, which shears a moving object. What counts is when rows form their image, not how long already-formed data takes to move.

From row timing to the error of your measurement

Speed multiplied by a row-time difference is a displacement between rows, not the error of every measured size. Take a rigid part crossing the frame at constant speed along the width: both boundaries of a width measured on one row shift equally, so subtracting their coordinates cancels the shift and shear alone leaves that width unchanged. A distance between points on different rows picks up the difference between their displacements, while an angle, a position or the shape of a contour is biased in its own way, all within the constant-velocity model rather than a camera test.

So carry that shift through the points themselves. Take the difference between the image-formation times of the rows your reference points fall on in the mode you will run and multiply it by the image velocity, which gives the displacement between those points rather than the error itself. Then move the coordinates of each point by its own displacement, recompute the width, the angle or the distance from the moved coordinates, and compare that change with the error share you allotted the shutter.

For a linear size the displacement across the whole region of interest, the sensor sub-area you read out, bounds that change from above: comfortably inside the share you allotted, row timing is not your problem, while exceeding it does not prove the task fails and staying under it does not prove the measurement accurate. An angle or the shape of a contour gets no such bound and needs its own recomputation.

What it looks like on the frame, and what else looks like it

Under uniform lateral motion and top-to-bottom row timing a vertical edge leans, because lower rows record the part at a later position; Allied Vision calls this shear. Under rotation or vibration, rows record different positions or orientations, so a straight spoke bends and the sequence wobbles, an effect colloquially called jello.

Shear is easiest to see when blur is small: image a straight edge that should be vertical across many rows, and one that stays sharp while tilting down the frame is evidence consistent with row-time displacement. The two can appear together, so a sharp edge does not rule out blur and a soft edge does not rule out shear.

Horizontal bands are a hypothesis rather than a diagnosis: Allied Vision warns that unstable pulse-width-modulated illumination can create that line structure, so repeat the acquisition under light you know to be stable, with exposure, gain and trigger timing fixed. If the bands vanish, or move in step with the light instead of the part, the lighting hypothesis holds.

When a rolling shutter passes on a moving part

A static scene is the easy case: still scenarios show no motion-induced geometric effect, Allied Vision states, so a stationary part under a stationary camera needs only steady light while the rows acquire. With motion there are two branches. Under stable continuous light, budget two quantities separately: the blur inside one row's exposure, and the geometric effect of the time difference between the rows carrying your reference points. A rolling shutter passes when both stay inside budget at the worst-case speed; lacking a documented flash window does not by itself disqualify it.

The other branch is a shared light pulse, which Basler recommends for fast rolling-shutter capture. Four conditions hold together: every row you need is light-sensitive at the same time, which Basler's Flash Window signal marks on models that document it; the pulse is short enough for the blur budget; delay and jitter keep it inside that interval; and ambient light leaves no second, smeared copy outside it. The exposure and strobe article carries that workflow.

Global Reset Release sits between the modes: rows start together but end or read with offsets, so without a flash or another common exposure end, brightness varies down the frame and a moving part still distorts. MV-CT0MU0060/R offers Rolling and Global Reset exposure modes (datasheet), a timing option to check against your speed rather than an answer. A global shutter removes the row-to-row term and leaves the blur budget standing, the distinction Basler's white paper draws.

Worked example: a rolling-shutter camera on a conveyor, blur only

MV-CA060-11GM is a rolling-shutter camera with 3072 × 2048 px at 2.4 µm (datasheet). The scenario assumes a field of 100 × 66.666667 mm at a working distance of 300 mm from the lens principal plane, parts travelling along the image width at 1000 mm/s, and a 1 px blur allowance.

Sampling, the pixels on each millimetre of the part, is 30.72 px/mm, so one pixel covers 32.552083 µm and the 1 px allowance caps the image-forming interval at 32.552083 µs (calculated). That ceiling lies inside the camera's exposure range of 27 µs to 2.5 s (datasheet), a comparison of ranges rather than a verdict on this scene. At 100 µs, an assumption of this example, the same calculation returns 3.072 px of blur (calculated).

Each figure above describes one row's exposure, so none says whether the frame is sheared. The second term reuses the sampling and adds a timing value the calculation never sees: image velocity is 30720 px/s (calculated), and the displacement between two rows is that figure times the difference between their image-formation times. Take that timing from the model documentation for the exact mode you will run, or send us the model and we confirm it.

Calculate for your case

The prepared link opens the calculator with the camera, both field dimensions, the working distance, the speed and the motion axis set. The 1 px allowance does not travel in the address and the page has no field for it, so reproduce the example through the optics route of the tools API: that request returns values.maxExposureUs of 32.552083 µs, the figure for 3072 px over 100 mm, or 30.72 px/mm.

The page itself shows 20 µs, a second calculation rather than a discrepancy: the address also carries a feature size of 0.1 mm and a target of 5 px, which the page turns into a requirement of 5000 px across 100 mm, or 50 px/mm, and at 1000 mm/s with a 1 px budget that gives 20 µs. One number is what the task demands, the other what the selected camera delivers, and neither models the shutter.

Equipment that fits this example

MV-CA060-11GM carries the example geometry: rolling shutter, 3072 × 2048 px at 2.4 µm (datasheet). MV-CT0MU0060/R adds Rolling and Global Reset modes on the same 3072 × 2048 px at 2.4 µm (datasheet). MV-CA050-12UM is the other branch, a global-shutter camera with 2448 × 2048 px at 3.45 µm (datasheet), where the row-to-row term does not arise. Mount, interface, cable and lens image circle remain separate checks on all three.

Common mistakes

• The calculator's exposure ceiling gets read as proof that the geometry is safe, so edges come back crisp while the measured dimension drifts with line speed; work out the row-timing term for your reference points before accepting the result.
• The frame period, one divided by the frame rate, gets used as the row-timing figure, or full-frame timing gets carried into a cropped region. A correction built on either leaves a residual lean that changes with the crop, since readout runtime can follow the height read, as Basler's A400 manual shows; take the timing from the model documentation for your region.
• A short pulse gets trusted because it is short. When it falls outside the interval in which the rows are sensitive together, the frame comes back dark over part of its height or with a brightness step across it; check the pulse, its delay and its jitter against the documented common-row window.
• Global Reset gets treated as a global shutter. Rows start together yet can end at different times, so brightness varies down the frame and a moving part still distorts, unless a flash or another mechanism ends the exposure in common.

When this rule breaks

The shear picture assumes one constant velocity along one axis for the whole acquisition. Acceleration, vibration and rotation put different rows at different positions or orientations, so the bias stops being a simple lean and no single skew factor corrects it. The same-row cancellation belongs to that model too, and says nothing about lens distortion, calibration or the algorithm that locates the edge.

Correcting the geometry in software needs four inputs together: the image-formation timing of the rows involved, the region read, the readout direction and the motion vector. That timing belongs to the configuration as much as to the camera, because the height of the region read can change it on cameras built that way, so confirm it from the model documentation for the region you will run; a figure confirmed for those conditions stays usable.

Next step

Settle it on the machine. Put a calibrated rigid target in the scene, move it at the worst-case speed and direction in the region you will read, and compare the measured dimensions against its known values. For a number beforehand, ask the camera documentation for the effective exposure start and end of the rows carrying your reference points and the difference between them, in the exact shutter mode, exposure, region height, resolution, frame rate and readout direction you will run; ask how the region changes it, and in the pulsed branch ask for the documented common-row or flash-window timing. Send us the camera model, the region of interest, the part speed and direction, and the dimension with its tolerance, and we confirm what the model documentation states.

Related

• Setting the exposure for a moving part, and deciding whether the light should be pulsed
• Deciding how many pixels a feature needs before you choose a camera

How this material was prepared

Prepared with MVisionPro AI agents from stated sources and the calculation core; MVisionPro retains editorial responsibility. A physical test or human engineering review is claimed only when explicitly stated. Read the editorial method.

Editorial status: verified. Content updated 2026-09-21.