Skip to content

Global shutter or rolling shutter

The same object, two cameras, two different images: one fan blade is whole, the other is bent. The difference is not the lens and not the exposure — it is the order in which the sensor reads its rows. This is the first question machine vision asks about a moving object, and the only one more light does not answer.

Exposure and shutter

Exposure time is the period from the shutter opening to it closing. During it, light falls on the photosensitive array, the photoelectric effect produces charge, and A/D conversion turns that into a grey value per pixel. At a given light level, a longer exposure means a brighter image: a long exposure shows the trajectory of a slow moving object, a short one records it accurately.

As film gave way to the chip, shutter control moved from mechanical to electronic. A new exposure begins once the photoelectric unit's charge is fully cleared, and ends when that charge is transferred out.

How each one works

In a global shutter sensor — CCD or CMOS — every pixel starts and stops exposing at the same instant. A rolling shutter CMOS sensor exposes one line at a time and reads each line out in turn; by the time it reaches the next line, the object has moved.

A complete exposure has four stages. Reset clears the charge in the photoelectric unit and opens the electronic shutter. Integrating is the exposure itself, where photoelectric conversion happens. Memory shifts the photoelectrons out of the cell. Readout transfers the pixel data.

The whole difference is whether the different lines of the frame are exposed at the same time. In a rolling shutter the readout wave sweeps down the sensor: the first line exposes first, the second one readout-time later, and so on. In exchange, each pixel cell needs only two transistors — less heat, lower noise, a simpler and cheaper structure. The price is paid on movement.

Where it shows

On a fast moving object, a rolling shutter distorts. The standard demonstration is a spinning fan: the global shutter image restores the blade's shape, the rolling shutter one deforms it, because each line caught the blade somewhere else.

Under varying light, horizontal stripes appear. A fluorescent lamp flickers at 50 Hz — a 10 ms period. At a 5 ms exposure each line falls into either the brighter or the darker half of that period, and because lines expose at different moments the frame comes out banded. With a global shutter every line begins and ends together, so there are no stripes.

The mechanism is clearest on a running dog: when the first line begins exposing, its head is just entering the frame; by the time the last line begins, the dog has nearly left. Each line caught it somewhere else, and the finished image shows a split dog.

What to do about it

If the speed is low and the light varies slowly, the difference is negligible. At high speed, the one fundamental fix is a global shutter sensor.

Where cost or noise forces a rolling shutter, a strobe mitigates it. The camera outputs a flash signal; while it is asserted, all lines are exposing at once, and within that window the frame is undistorted.

Three caveats about the strobe. The signal is not output for every exposure: if the exposure is too short and the readout too long, the lines never overlap and no signal appears. The flash must be shorter than the exposure. And when the signal lasts microseconds, not every strobe can switch fast enough to catch it.

See the global shutter cameras in the catalogue →
Global shutter or rolling shutter | innoSMART Automation