Motion Blur Photography and Shutter Speed

What is Motion Blur?

Motion blur is related to exposure. The object's image moves across the sensor while the shutter is open, and that motion smears the object over every pixel it crossed: linear blur when the motion is a translation, angular blur when it is a rotation, and both at once when it is both.

Size it before you buy a lens or a light. The smear at the sensor is roughly v × t × m: object speed times integration time times optical magnification. Divide by the pixel pitch for the answer in pixels.

Take a part moving at 1 m/s under a lens working at 0.05x magnification, captured with a 500 µs exposure onto a 3.45 µm pixel. The object travels 0.5 mm, the image smears 25 µm, and that is about 7 pixels. Drop the exposure to 100 µs and the smear falls to 1.4 pixels. Whether that is small enough depends on your measurement tolerance, not on a general rule.

Integration-time blur is not the same defect as rolling-shutter skew. Blur comes from the object moving during one pixel's exposure window. Skew comes from the sensor starting each row's exposure at a different moment, which leans vertical edges over and wobbles the whole frame when the camera vibrates. The fixes differ: shorten the integration time and add light or a strobe for blur; for skew, move to a global-shutter sensor or strobe inside the window where every row is integrating at once.

In a dark enclosure the flash duration, not the shutter setting, is what actually stops the motion.

What does Motion Blur do?

Motion Blur makes objects in images look blurry, for both humans and for computers.

Exposure Has a Major Impact on Computer Vision and Machine Vision

Pei et al. show how motion blur degrades the performance of image classification. They show that VGG and AlexNet are more affected by angular motion blur than blur length, but both jointly degrade performance.

Avoiding motion blur matters wherever the scene or the camera moves and the task reads an edge: barcode and OCR decoding, defect detection, and dimensional measurement. A stationary part under steady light does not need it.

References and Related Links

Pei et al. "Effects of Image Degradations to CNN-based Image Classification"