Global vs Rolling Shutter for Machine Vision
Choose a sensor shutter architecture by separating within-row blur from whole-frame geometric distortion.
A rotating fan blade is sharp in every local region but bends across the image. Shortening exposure helps only slightly. The problem is not conventional motion blur alone: different sensor rows were exposed at different times.
Global and rolling shutters can both produce excellent industrial images, but they fail differently. The choice depends on motion, illumination timing, frame readout, sensitivity, and the geometry your algorithm must preserve.
What you will learn
Understand how global and rolling shutters expose a frame.
Separate per-row motion blur from row-to-row distortion.
Calculate the top-to-bottom time skew of a rolling sensor.
Evaluate synchronized pulsed-light alternatives.
Validate shutter choice on real motion and lighting.
Technical foundation
Global shutter
A global-shutter sensor starts and ends exposure for all pixels at approximately the same time, then reads the stored signal. Moving geometry is captured at one common time, subject to the chosen exposure duration and sensor implementation.
This is usually the straightforward choice for fast conveyors, measurement of moving objects, robot motion, vibration, and strobed inspection.
Rolling shutter
A rolling-shutter sensor exposes rows at staggered times. Each row can have a short exposure, yet the last row begins later than the first. Motion during this row-to-row interval creates skew, stretch, compression, wobble, or inconsistent illumination.
Rolling sensors can offer advantages in cost, sensitivity, read noise, resolution, or availability. They work well when the scene is static, motion is slow relative to line timing, or illumination and sensor modes support controlled simultaneous exposure.
Two different motion calculations
Within-row blur depends on exposure time:
Blur distance = object speed × exposure time
Whole-frame distortion depends on the time difference between rows:
Top-to-bottom skew time ≈ number of row intervals × line time
The object motion during that skew interval can be much larger than its motion during one row’s exposure.
Pulsed illumination and pseudo-global modes
Some rolling-shutter cameras provide a period when all rows are simultaneously sensitive. A short strobe inside that overlap can create a global-like effective exposure, usually with frame-rate or timing trade-offs. This is sensor- and mode-specific; confirm the timing diagram rather than assuming any rolling sensor supports it.
[Suggested visual: global and rolling shutter exposure timing]
Purpose: Explain why a short exposure does not eliminate row-time distortion.
Required elements: Sensor rows versus time, global rectangular exposure, diagonal rolling windows, and a short all-row strobe in an overlap region.
Suggested caption: “Exposure duration controls local blur; row timing controls rolling-shutter geometry.”
Accessible alt text: “Timing plots compare all sensor rows exposing together with rows exposing sequentially from top to bottom.”
Engineering workflow
1. Describe all relative motion
Include conveyor translation, rotation, robot motion, camera vibration, part oscillation, and flickering illumination. Record maximum speed and direction in image space.
Why it matters: rotational features can move much faster at their outer radius than the part centre.
2. Define geometric tolerance
Decide whether the task tolerates skew. A presence check may accept it; circle measurement, pose estimation, metrology, and code geometry may not.
3. Obtain real sensor timing
Find exposure mode, row or line time, frame readout time, overlap behaviour, trigger delays, and illumination output timing for the exact camera configuration. ROI height, bit depth, and readout mode may change timing.
Common failure: selecting by the word “CMOS” without checking shutter architecture and mode.
4. Calculate both blur and skew
Use exposure time for local smear and row-time span for whole-frame displacement. Convert both into object units and pixels.
5. Evaluate lighting compatibility
For a global shutter, align a short pulse with the exposure-active window. For a rolling sensor, determine whether all rows overlap and whether the pulse can illuminate them uniformly. Ambient light during the rolling exposure can still contribute distortion even if a strobe is used.
6. Compare signal performance
Do not choose shutter architecture in isolation. Compare quantum efficiency, read noise, full-well capacity, dynamic range, pixel size, frame rate, and required illumination using comparable data—preferably EMVA 1288-compliant specifications where available.
7. Prototype the worst motion
Use real parts, speed, rotation, exposure, strobe, and trigger timing. Check geometry across the frame and repeat at speed extremes.
Worked example: short exposure, large rolling skew
Hypothetical application: a 2,048-row rolling-shutter sensor has a 10 µs row time. Conveyor speed is 500 mm/s. Each row uses a 50 µs exposure.
Approximate top-to-bottom time skew:
2,048 rows × 10 µs/row = 20.48 ms
Object movement during the frame skew:
500 mm/s × 0.02048 s = 10.24 mm
Local movement during each 50 µs row exposure:
500 mm/s × 0.000050 s = 0.025 mm
Each row can therefore look locally sharp while the whole object is displaced by more than 10 mm between top and bottom row timing. A global-shutter frame with the same 50 µs exposure limits the whole-frame motion interval to approximately 0.025 mm, excluding other timing effects.
If the rolling sensor has a verified all-rows overlap mode, a sufficiently short synchronized strobe may reduce effective geometric distortion. The ambient contribution must be suppressed, and frame-rate/trigger constraints must be measured.
[Suggested visual: moving circular part captured by both shutters]
Purpose: Distinguish blur from shape distortion visually.
Required elements: True circle, globally captured circle with slight uniform blur, rolling-shutter slanted ellipse, direction arrow, and timing annotations.
Suggested caption: “Rolling-shutter artifacts can preserve local sharpness while corrupting global geometry.”
Accessible alt text: “A moving round part remains round with global shutter but appears slanted and stretched with rolling shutter.”
Shutter decision table
| Application condition | Global shutter | Rolling shutter |
|---|---|---|
| Fast translation or rotation | Strong default | Requires timing proof or pulsed-light strategy |
| Precision geometry on moving part | Preferred | High risk of skew |
| Static scene | Works | Often suitable |
| Very low light | Compare real sensor performance | May offer sensitivity advantages depending on sensor |
| Synchronized short strobe | Straightforward timing | Possible only with compatible overlap/mode |
| Flickering illumination | Common-time exposure helps | Rows may see different light phases |
| High resolution/cost constraint | Product-dependent | Often broader choices |
| Robot or vibration motion | Strong default | Validate row-time distortion carefully |
Common mistakes
Believing a short exposure eliminates rolling artifacts. It reduces within-row blur, not row-to-row time skew.
Comparing only frame rate. Row time and readout mode determine distortion.
Assuming every global shutter has identical image quality. Sensor noise, sensitivity, and dynamic range still differ.
Assuming any strobe makes a rolling shutter global. All-row overlap and ambient suppression must be verified.
Testing translation but not rotation. Rotating edges can reveal severe distortion.
Ignoring LED or mains flicker. Different rows may receive different illumination levels.
Using a reduced ROI without rechecking timing. ROI can change readout and overlap behaviour.
Evaluating only visual sharpness. Measure geometry, scale, and algorithm output across the frame.
Validate under production conditions
Test the final sensor mode at maximum linear and angular speed, acceleration, vibration, exposure, ROI, bit depth, and lighting state. Use geometric artifacts or features that reveal skew, not only texture.
Record object-space smear, top-to-bottom displacement, measurement bias, code-read rate, pose error, and false decisions. Repeat under ambient-light extremes and strobe timing limits.
If using a rolling sensor with pulsed illumination, capture timing traces for trigger, row exposure or all-row window, and light pulse. Verify behaviour after camera firmware or acquisition-mode changes.
Key takeaways
Global shutters expose the frame at a common time; rolling shutters stagger rows.
Exposure time controls local blur, while line timing controls whole-frame skew.
A rolling frame can look sharp and still be geometrically wrong.
Synchronized pulsed light can help only when the exact sensor mode supports it.
Compare timing and image quality for the complete camera, not shutter labels alone.
Follow this Hashnode blog for more practical camera-selection guidance, and connect with Kivanc Ekici on LinkedIn. For related machine-vision and automation information, visit ITAGE.
Frequently asked questions
Is global shutter always better for machine vision?
No. It is usually easier for moving geometry, but a rolling sensor may be suitable for static scenes or controlled pulsed-light modes and may offer other performance advantages.
Does rolling shutter cause motion blur?
It can cause both local exposure blur and row-to-row geometric distortion. These are separate effects.
Can a strobe fix rolling shutter?
Only if the sensor provides a verified interval when all required rows are sensitive and ambient light is controlled. Check the exact timing mode.
How do I calculate rolling-shutter distortion?
Multiply the time difference between relevant rows by the object’s image-plane velocity, then convert the displacement into pixels or object units.
Why does a circle appear oval or slanted?
Different rows captured the moving circle at different positions. The effect can be rolling-shutter skew rather than lens distortion.