Cameras are the most common drone payload and the easiest to buy badly, because spec sheets advertise pixels while missions live on resolvable detail per situation. Here's how to map mission requirements to sensors, optics and gimbals — and integrate them so the data holds up.

Requirements before catalogues

Write these three numbers down first; they select the camera for you:

  1. GSD or DRI: for mapping/inspection, required ground sample distance (cm/pixel) at working distance; for security/SAR, detection-recognition-identification ranges.
  2. Standoff distance: how close the aircraft may fly (regulation, safety, turbulence, wake from the asset).
  3. Product: stills for photogrammetry, live video for operators, radiometric data for analysis — each drives different silicon.

GSD arithmetic: GSD = (pixel pitch × distance) / focal length. A 1-inch 20 MP sensor (2.4 µm pitch) with a 24 mm lens at 60 m gives ~0.6 cm/px — bolt this into mission planning and the survey guide's workflows.

EO (visible) sensors: size beats megapixels

Sensor format determines light-gathering and dynamic range; megapixels on a tiny sensor mostly buy noise. Working hierarchy: 1/2.3" (consumer, daylight-only), 1" (the commercial sweet spot), M4/3 and APS-C (premium mapping/cinema). Two features matter more than resolution for moving platforms:

  • Global or fast-readout shutter — rolling shutter plus vibration equals jello imagery no software fully repairs; for photogrammetry it corrupts geometry, not just looks. Mechanical shutters remain the mapping standard.
  • Real exposure control via API — auto modes tuned for tourists make inspection decisions (locked exposure across a survey, spot metering on the asset) impossible.

Thermal: buy radiometry, mind the physics

Uncooled LWIR microbolometers (640×512 the professional floor, 1280-class arriving) cover most commercial work. The decisions:

  • Radiometric vs not — see FAQ; for inspection, radiometric, full stop.
  • Lens choice is permanent — thermal zooms are exotic; most cores fix focal length. Pick from required spot size: measuring a 5 cm defect wants that defect covering multiple pixels at standoff.
  • Physics caveats your ops team must know: emissivity (shiny metal lies), reflected-sky temperature, sun-soak vs internal-heat ambiguity (solar panels), and NETD noise in cold scenes. Radiometric data plus untrained interpretation is how customers get confidently wrong reports.
  • Export control: higher-performance thermal cores carry dual-use export restrictions (frame rate and resolution thresholds) — a supply and shipping workstream, not a footnote, if you sell internationally.

Gimbals: the spec that matters is milliradians

Stabilisation residual determines usable focal length: at 0.1° (1.7 mrad) residual, a 10° FOV telescope smears; at 0.02° it's sharp. Evaluate with your camera mass installed — gimbal performance is a tuned-mass problem, and a gimbal rated "up to 800 g" behaves differently at 790 g than the demo did at 300 g. Also check: slew rate for target tracking, angular range (does it see straight down? backwards?), slip-ring vs wrapped cables for continuous pan, and the control API's latency for closed-loop tracking from the companion computer.

Integration pitfalls that ruin data

  • Vibration through the isolators: gimbal isolation tuned for one payload mass resonates at another — see the mechanical section of the integration checklist.
  • No time sync: imagery without PPS-disciplined timestamps can't be precisely georeferenced — mandatory for mapping, priceless for inspection audit trails.
  • Downlink mismatch: a 4K sensor behind a 2 Mbps video link shows the operator mush while recording gold; design the record-vs-stream split explicitly.
  • Heat: zoom blocks and encoders throttle in sealed pods; thermal cores drift with their own temperature (radiometric accuracy specs assume thermal equilibrium — allow warm-up time in ops procedures).
Dual-sensor pods

EO + thermal in one pod is the inspection default for good reason: the EO context makes thermal findings interpretable and reportable. Check that the two sensors are boresight-aligned and simultaneously triggered — "aligned in marketing" and "aligned in metadata" are different products. Feeding these streams into onboard analytics is covered in the AI drone integration guide.

Frequently asked questions

What is the difference between radiometric and non-radiometric thermal cameras?

A radiometric thermal camera measures calibrated temperature per pixel, so imagery supports quantitative analysis (a hotspot is '84 °C', not 'bright'). Non-radiometric cores show relative contrast only. Inspection and solar/electrical work almost always justify radiometric.

What gimbal specification actually matters for image quality?

Stabilisation residual in milliradians (or the practical proxy: sharp imagery at your longest focal length in wind), not the marketing axis count. A gimbal holding ±0.02° supports far more zoom than one at ±0.1°, whatever both boxes claim.

Why are my drone photos blurry even with a gimbal?

Usual order: rolling-shutter distortion from vibration (balance props, isolate properly), shutter speed too slow for platform motion, gimbal tuned poorly for the payload mass, or focus/IR-focus shift on thermal optics. Fix vibration at the source first.