Your camera sits at 99 fps and no setting moves it. The number itself is the diagnosis: 1000 / 99 is a 10.10 ms frame period, and about 10 ms is exactly where an auto exposure settles when the scene is not bright enough. A sensor cannot begin exposing the next frame until the current exposure has ended, so the shutter time is a hard ceiling that no frame-rate request can lift. Here you will read the ceiling off your own numbers, test which limit you have hit — exposure, readout or a rescaling pipeline — and work out how much light the shorter shutter needs. The math is the same for a GigE area scan camera on a line, a USB3 board camera in a compact fixture, or a triggered inspection shot.
The ceiling: the exposure must fit inside the frame period
The frame period is 1,000 divided by the frame rate, in milliseconds. The exposure has to finish inside that period, because the sensor cannot start the next exposure until the current one ends. Whichever ends last — the exposure or the sensor's readout — sets the ceiling, so the useful form of the rule is:
Read the middle line as the shutter's own limit: shorten the exposure and the ceiling rises in exact inverse proportion.
| Exposure time | Ceiling from the exposure alone |
|---|---|
| 10 ms | 100 fps |
| 8 ms | 125 fps |
| 5 ms | 200 fps |
| 2 ms | 500 fps |
| 1 ms | 1,000 fps |
| 100 µs | 10,000 fps |
The second ceiling, readout, belongs to the sensor mode: lines multiplied by line period. On a hardware trigger the two terms add instead of taking the larger one, which is why triggered rates are quoted as exposure plus readout — the maximum frame rate on a hardware trigger guide works that case.
Why the reading is exactly 99 fps
Auto exposure targets a brightness; when the scene is short of light it stretches the shutter until it reaches the largest value that still fits in a frame, which is one frame period. At a 100 fps request that is 10 ms. The pipeline adds a fraction of a millisecond — readout tail, transfer, processing — and the camera lands at 1000 / 10.10, about 99 fps.
Two things follow. Requesting a higher frame rate does not shorten the shutter: the exposure loop simply puts the light back. And lowering the output resolution changes nothing at this stage: the exposure and the sensor scan are untouched, so the period stays 10.10 ms.
The same mechanism appears as a rate that halves when auto exposure priority takes over — see auto exposure priority: why your frame rate drops to half — or drifts down as ambient light falls, as in why your GigE camera frame rate drops at longer exposure times.
Test the three limits in this order
Three different limits produce a stuck fps: the exposure, the readout, and a pipeline that rescales instead of resizing. Each test below takes a minute.
- Read the exposure the camera is actually using, from frame metadata or the exposure readback, while it streams at the stuck rate. If it sits at the frame period — 10.1 ms at 99 fps — the shutter is the ceiling. The read exposure time from frame metadata under auto exposure guide shows the readback technique.
- Compute 1000 / exposure and compare it with the measured fps. Equal within a few percent: the shutter is the limit. Far below: readout or pipeline is the real ceiling.
- Disable the auto functions and set a fixed exposure under the frame period. The fps should now follow 1000 / exposure until it flattens; the flat number is your readout ceiling. If the rate only recovers once the autos are off, that was the cause — see camera fps lower than the datasheet: turn off auto functions.
- Set a smaller output size and re-measure. If the fps does not move, the pipeline rescales frames after capture while the sensor still scans the full array; only a different sensor mode changes the readout time.
Worked example: from 99 fps measured to 120 fps needed
Inputs, each with its source:
- Measured frame rate: 99 fps — the stuck reading in the stream.
- Exposure from the metadata readback: 10.0 ms — auto exposure settled at one frame period.
- Target: 120 fps — the next mode the sensor offers.
Rule in words: the exposure must fit inside the target period, and the photons per frame fall in proportion to the shutter.
Arithmetic, step by step:
- Measured period: 1000 / 99 = 10.10 ms.
- Exposure ceiling at the current shutter: 1000 / 10.0 = 100 fps. The reading is 99, which is 1% under the ceiling — that gap is readout and pipeline overhead: 100 / 99 = 1.01.
- Target period: 1000 / 120 = 8.33 ms, so the exposure must be 8.33 ms or less.
- Set the shutter to 8 ms: new ceiling 1000 / 8 = 125 fps, which is 4% of headroom over the target: 125 / 120 = 1.04.
- Light budget: photons per frame drop to 8 / 10 = 0.8 of what they were, a 25% loss worth log2(1.25) = 0.32 stop. A third of a stop of aperture, or about 2 dB of gain (20 × log10(1.25) = 1.9 dB), buys it back.
- The 1 ms counter-case: the ceiling becomes 1000 fps but the light falls 10x, and log2(10) = 3.3 stops — a cost you only accept when motion blur, not the fps target, forces it.
Decision: set the exposure just under the target frame period — 8 ms for 120 fps — and recover the 25% light with aperture or a slightly brighter lamp. Reserve sub-millisecond shutters for frozen motion, and pay for them with a strobe, not gain.
The fix: set the shutter, then buy back the light
Settings first. Cap the auto exposure below the frame period or go fully manual, fix the gain, and verify the values actually loaded after boot — the cap the maximum auto exposure time to stop motion blur guide covers the cap itself, and auto exposure upper limit not applied: check startup settings covers the case where the cap never takes effect. Read the exposure back from metadata once the line runs: a setting that never loaded looks identical to one that did, until you check.
Photons second. Every halving of the shutter halves the photons per frame. Buy them back in this order: aperture (each f-number step is a factor of two), LED current or a larger emitting area, a strobe pulse packed inside the exposure window, and gain last, because it amplifies noise with the signal.
For moving parts a strobe is the clean answer: fire a pulse equal to or shorter than the exposure and motion freezes at the pulse width. A 48 V high power strobe light of 84 x 60 x 60 mm, such as the LS-SPC-0129, is built for exactly this duty on fast lines.
- Size
- 50-100 mm
- Voltage
- 48 V
- Type
- Strobe
One safety check belongs here: strobes driven hard, and any infrared lamp, at close working distances fall under the IEC 62471 photobiological limits — check IR LED eye safety at close range: IEC 62471 limits before mounting a lamp inside the operator's line of sight.
When the shutter is short and the fps still will not move
Once the exposure sits well under the period, 1000 / exposure no longer describes the camera; the readout does. The levers are then at the sensor, not in the settings: a reduced active region, binning or skipping where the sensor offers them, or a different mode with a shorter line period. An output-only rescale touches none of these — which is why a smaller video size left the fps untouched in test 4.
On a triggered camera the budget is stricter: each frame spends exposure plus readout. A 5 ms exposure on a mode whose readout takes 12 ms, for example, yields at most 1000 / (5 + 12) = 59 fps no matter how short you set the shutter. The compute camera frame time from exposure and readout walkthrough handles that sum, and the same arithmetic sits behind triggered frame rate troubleshooting on industrial cameras.
If the inspection needs hundreds of fps at full resolution, buy the mode, not the hope. A 3.1 MP USB3 board camera with a 1/1.8" global shutter and a 120 fps mode, such as the LS-BCM-0003, fits compact builds where a rescaling pipeline would otherwise eat the rate.
- Sensor format
- 1/1.8"
- Lens mount
- C-mount
- Interface
- USB3
- Resolution
- 5 MP
At 120 fps the period is 8.33 ms, so an 8 ms exposure fits with 0.33 ms to spare — the same arithmetic as the worked example above. For judging whether a compact camera holds its rate inside your fixture, the board camera evaluation checklist for inspection systems lists the checks that matter before you order, and the guide on how to select a high-speed camera for 5,000 fps puts the readout term first when the targets get extreme.
Checks that keep the fps where you set it
- Set the exposure under the target period with margin — 8 ms against an 8.33 ms period — and read it back after every boot; the camera startup frame rate: verify saved settings and readback guide covers the readback checks.
- After any pixel format, ROI or binning change, measure the fps again: linked attributes move more than one value at a time, and the changing pixel format halves your frame rate: check linked attributes guide shows the classic halving.
- Budget the lighting for the shortest exposure your motion spec needs before ordering the lamp — the light finder and the special lights category, which holds the strobes, sort by that duty.
- Keep auto functions off on a fixed line; where ambient light genuinely varies, use a capped auto exposure and confirm at startup that the cap applies.
Next step: match camera and light to the exposure you have budgeted. The camera finder filters area scan cameras and board-level cameras by frame rate, shutter type and resolution; the configurator builds the chain end to end. If the numbers fight — exposure, motion blur and light budget pulling in different directions — the contact page is the fastest way to get a second pair of eyes on the arithmetic, and more fixes like this sit on the cameras topic page.
Questions
Why does my camera sit at exactly 99 or 100 fps?
Auto exposure stretches the shutter toward one frame period when the scene is short of light: 10 ms at a 100 fps request. The pipeline adds roughly 0.1 ms of overhead, so the camera lands at 1000 / 10.1, about 99 fps. Raising the requested frame rate does not shorten the shutter; only a shorter exposure, with light behind it, lifts the ceiling.
How much light do I lose when I shorten the exposure, in stops?
Light per frame scales with the shutter: 10 ms down to 8 ms is a 25% loss, worth 0.32 stop; 10 ms down to 1 ms is a 10x loss, worth 3.3 stops. Aperture recovers it at one stop per f-number step — f/4 to f/2.8 is one stop — but every stop wider costs depth of field, which is why added LED light or a strobe is usually the better trade.
Does lowering resolution increase the frame rate?
Only when the sensor's scan area shrinks. An output-only rescale — what many ISP pipelines do when you pick a smaller video size — leaves the sensor scanning the full array, so the readout time and the 99 fps behaviour stay exactly as they were. To gain fps from resolution, select a sensor mode with a reduced active region, or binning or skipping if the sensor supports it.
What exposure should I set for a 120 fps line?
Set 8 ms against the 8.33 ms period: the exposure ceiling lands at 125 fps and about 0.3 ms of the period stays free for readout overhead. Going to 1 ms buys nothing unless motion blur demands it — it burns 90% of your light for speed the readout will not deliver.
Why did the image blow out or turn noisy after I shortened the shutter?
With auto functions still enabled, the pipeline re-balances gain against the new shutter frame by frame and can overshoot into artifacts. Fix both values manually, disable auto exposure, re-test, and judge the brightness only then. If the image is now too dark, add light rather than gain: gain raises the noise floor with the signal, so the darker parts of the image go grainy first.
Can a hardware-triggered camera be made faster by shortening exposure alone?
No. On a trigger each frame costs exposure plus readout, so once the exposure is shorter than the readout, cutting it further changes nothing — the mode's readout time is the ceiling. The lever is then a faster sensor mode, a reduced region, or a camera chosen for frame rate at your resolution.

