Can a stationary subject have motion blur?

Yes. Camera movement during exposure can move the subject's image across the sensor.

Does a global shutter prevent motion blur?

No. Simultaneous exposure across the sensor avoids one source of geometric distortion, but movement during the exposure can still smear detail.

Can software always recover a sharp original?

No. Recovery is limited by lost information, noise, and modeling assumptions; a convincing reconstruction is not automatically an accurate one.

AI Glossary / Graphs and symbolic AI / Beginner

Motion blur

Definition

Motion blur is the smearing of image detail caused by movement during the time an image sensor collects light. The movement can come from the subject, the camera, or both. A feature contributes light across several image positions during the exposure instead of appearing at one sharply defined position.

What it means

A camera frame is not always an instantaneous snapshot. During its exposure interval, a moving edge may sweep across multiple pixels. The recorded image combines information from that movement, making text, boundaries, or small features harder to distinguish.
Motion blur matters to AI because a vision model receives the recorded image, not the original sharp scene. Recognition, tracking, measurement, and inspection can become less reliable when important detail is smeared. Blur can also be intentional in photography or rendering, so whether it is undesirable depends on the task. This entry concerns capture-related motion blur rather than display persistence effects.

How it works

For approximately constant motion in the image plane, blur extent can be estimated as image speed in pixels per second multiplied by exposure time in seconds. Faster image motion or a longer exposure increases the distance a feature travels while light is collected.
Shorter exposure can reduce the smear but also collects less light under otherwise identical conditions. Illumination, aperture, sensor behavior, and noise therefore matter. Raising gain brightens the readout while amplifying noise as well; it does not recreate lost spatial detail. A global shutter avoids row-timing differences but does not eliminate motion blur during each exposure.

Three practical examples

A conveyor inspection system can lose fine defect detail when parts move too far during exposure, even if the camera captures many frames each second.
A mobile document scanner can produce blurred lettering when the device moves while the image is being captured.
A sports-tracking system can struggle to localize a fast object when its image becomes a streak rather than a compact shape.

Think of it this way

Think of drawing a bright dot while the paper moves underneath it: the mark spreads into a line instead of remaining a point. The analogy is limited because cameras integrate light, blur can follow complex motion, and different parts of a scene may move differently. Real blur is not always one uniform straight-line smear.

Worked example

Suppose an invented conveyor setup maps one millimeter of object motion to four pixels in the image. A part moves sideways at 250 millimeters per second, approximately parallel to the image plane. Its image speed is therefore 250 times 4 = 1,000 pixels per second.
At an exposure time of 5 milliseconds, or 0.005 seconds, estimated smear is 1,000 times 0.005 = 5 pixels. At 1 millisecond, it is about one pixel. At 0.5 milliseconds, it is about half a pixel. These estimates assume steady motion and approximately constant magnification over the relevant region.
If the camera records 100 frames per second, frames begin roughly 10 milliseconds apart, but that interval does not determine the exposure duration by itself. A 5-millisecond exposure can still smear by five pixels at that frame rate. Reducing exposure requires checking whether enough light remains for the inspection task; the arithmetic alone does not specify a complete camera setup.

Illustrative case study

In a fictional packaging line, an OCR model performs well on stationary labels but makes errors on moving products. The team initially considers retraining the model. A capture review shows that narrow letter strokes smear across several pixels at the current exposure.
Engineers compare shorter exposures with suitable lighting and collect matched moving-label tests. They evaluate recognition performance and noise, then include realistic residual blur in model testing. The scenario illustrates investigating image formation before attributing every error to the model. It does not claim that a particular camera setting guarantees accurate OCR or that a measured improvement occurred.

Common misconceptions and mistakes

Do not confuse motion blur with defocus blur. Refocusing a lens will not necessarily remove smear caused by movement. Do not assume a higher frame rate automatically means a short enough exposure, or that a global shutter freezes all motion.
Rolling-shutter distortion is another distinct effect: different rows capture the scene at different times, which can bend or skew moving shapes. It can coexist with motion blur. Finally, AI deblurring may produce plausible-looking detail without reliably recovering the exact original scene. Restored images should be evaluated against the actual task rather than judged solely by visual sharpness.

Business use

For business vision projects, motion blur belongs in the acquisition requirements and acceptance tests. Specify the smallest feature that must remain usable, the expected image motion, exposure limits, and lighting conditions. Test representative fast motion as well as convenient stationary samples.
Keep training and evaluation conditions aligned with deployment. A model trained only on crisp images may fail when motion or low light changes. Conversely, tolerable blur depends on the application: identifying a large object and reading tiny print require different detail. Use task performance and measurement accuracy to set tolerances rather than treating one pixel as a universal pass/fail law.

Check your understanding

In the worked setup, what exposure gives an estimated two-pixel smear?

Answer: Divide two pixels by 1,000 pixels per second to obtain 0.002 seconds, or two milliseconds. This uses image speed, not simply the object's speed without accounting for magnification.

Related terms

Image noise
Computer vision
Optical flow
exposure time
shutter speed
rolling shutter
global shutter
defocus
deblurring
image acquisition

Learn next

Double the magnification in the worked example while retaining the same object speed and exposure. The image speed doubles, so the estimated smear doubles too. Then compare a sharp but noisy image with a smoother blurred image using an actual recognition or measurement task to understand the acquisition tradeoff.

Sources

Sources: Basler, Global Shutter, Rolling Shutter—Functionality and Characteristics of Two Exposure Methods, section 5: https://assets-ctf.baslerweb.com/dg51pdwahxgw/7ntd9P0LbEeEmxKj6Q6J1h/cb845234f745dddab95f3ee64994dd7d/BAS1401_White_Paper_Rolling_Shutter.pdf . Basler product documentation, exposure and gain: https://docs.baslerweb.com/optimizing-image-quality . Conveyor calculations and the packaging case are original examples.

Previous: Motion analysis Next: Motion capture