An uncooled microbolometer has no absolute reference. Every pixel responds slightly differently, and that response drifts with the detector’s own temperature. Everything the calibration chain does is an attempt to hold a moving target still long enough to produce a number you can trust.
Key takeaways
- Uncooled microbolometers drift with their own temperature, which is why shutter-based correction is unavoidable rather than a design shortcut.
- The correction chain has distinct stages, and only some of them can be refreshed in flight.
- Every shutter click costs you a fraction of a second of blind time — an issue for tracking and for any moving target.
- Shutterless operation does not remove the physics; it moves the correction into software and accepts the residual error.
On this page
Why calibration is unavoidable
Each microbolometer pixel is a tiny thermally isolated resistor. Manufacturing variation means no two have identical resistance or identical response slope. Uncorrected, the raw image is dominated by fixed-pattern noise that swamps the scene.
Worse, the response changes as the detector substrate warms. An uncooled core with no thermal stabilisation can drift several kelvin of apparent temperature during the first ten minutes after power-on, and again whenever ambient conditions change. A camera that reads 62 °C at minute two and 58 °C at minute twenty has not seen the target change — it has drifted.
The four stages of the chain
| Stage | What it corrects | When it runs | Reference used |
|---|---|---|---|
| Factory NUC | Pixel-to-pixel gain and offset variation | Once, at manufacture | Two blackbodies at known temperatures |
| Two-point calibration | Absolute radiometric accuracy | Factory; recheck annually | Traceable blackbody sources |
| Flat-field correction (FFC) | Accumulated offset drift | Every 1–5 min, or on ΔT trigger | Internal shutter as uniform scene |
| Shutterless correction | Drift, without a mechanical shutter | Continuously | Scene statistics + substrate temperature model |
| Ambient compensation | Housing and optics self-emission | Continuously | Internal temperature sensors |
| User emissivity / distance | Scene-dependent radiometric error | Set per target | Operator input |
What the shutter click costs you
FFC closes an internal shutter, presents the detector with a uniform surface, and re-learns each pixel’s offset. It takes 0.3–1.5 seconds, during which the video freezes on the last good frame. On an inspection flight that is invisible. On a search flight, a one-second freeze at the wrong moment is a target you did not see.
Most payloads let you set the interval, defer FFC manually, or trigger it on demand. Good practice on time-critical missions is to force an FFC immediately before the run-in and then suppress automatic cycles for the duration of the pass.
Shutterless designs avoid the freeze by modelling drift from substrate temperature and scene statistics. They are excellent for continuous surveillance and generally less accurate radiometrically than shutter-based correction — a genuine trade, not a marketing distinction.
What still drifts after all of this
- Warm-up. Expect 5–15 minutes before radiometric readings stabilise. Measurements taken during climb-out are the least trustworthy of the flight.
- Window and optics self-emission. A sun-heated germanium window emits into the detector. Ambient compensation models it; direct sun on the lens defeats the model.
- Emissivity assumptions. The largest single error source in practice, and one no internal calibration can fix. See emissivity in drone thermography.
- Atmospheric path. At long range, the air between camera and target both attenuates and emits. Payloads with laser ranging can compensate automatically; those without rely on an operator-entered distance.
- Ageing. Detector response changes slowly over years. Annual verification against a reference source is the only way to catch it.
How this shows up in our payloads
Radiometric models such as OP-90A and OP-125A pair shutter-based correction with a laser rangefinder, so atmospheric path compensation uses a measured distance rather than an operator guess. That single link removes the most common long-range measurement error we see in the field.
Related reading
Technology: radiometric temperature measurement and the thermal image processing pipeline.
Field practice: radiometric vs non-radiometric, seven thermal imaging mistakes and cold-weather operations.
What to verify before you trust a temperature reading
A radiometric payload will always give you a number. Whether that number is defensible depends on a short list of conditions that no automatic correction can check for you.
The single largest error in practice is emissivity, and it is entirely under operator control. The second largest is reflected apparent temperature — a cold sky reflecting off a shiny surface pulls the reading down, a nearby hot object pushes it up. The third is atmospheric path, which matters increasingly beyond about 100 m and which payloads with integrated ranging can compensate automatically.
None of these are exotic. They are the standard conditions of any thermographic measurement, and the reason certified thermography training exists. What changes on a drone is that you cannot walk up and check the surface, so the discipline has to be front-loaded into the flight plan.
- Warm-up complete: at least 5–15 minutes powered before the first measurement.
- Emissivity set for the actual material, not left at the 0.95 default.
- Reflected apparent temperature entered, especially for shiny surfaces under open sky.
- Distance entered or measured by laser, for any target beyond about 100 m.
- Viewing angle within about 45° of normal to the surface.
- A recent FFC, and no FFC pending mid-measurement.
- Where possible, a comparative reference in the same frame — comparison beats absolute measurement almost every time.
FAQ
Why does my thermal image freeze for a second every few minutes?
That is flat-field correction. An internal shutter closes to give the detector a uniform reference and re-learn each pixel’s offset. It is required to hold image quality and radiometric accuracy on an uncooled core; most payloads let you defer or manually trigger it during critical passes.
How long should I let the payload warm up before taking measurements?
5–15 minutes. Uncooled detectors drift measurably while the substrate temperature stabilises, and readings taken during climb-out are the least reliable of the sortie.
Do shutterless cameras measure temperature accurately?
Less accurately than shutter-based designs, generally. They trade some radiometric precision for uninterrupted video, which is the right trade for continuous surveillance and the wrong one for inspection work that produces numbers.