Thermal Camera Range Estimator with Visualization

These are first-order estimates for scoping and comparison, not a performance guarantee. Real detection depends on scene clutter, sun-heated surfaces, image processing, operator attention and how the flame actually radiates. Treat the signature-limited figure as a ceiling under favourable conditions.

Signature-limited range

A fire is far hotter than its surroundings, so it can be detected long after it has shrunk below one pixel. What matters is whether it lifts the apparent temperature of the pixel it lands in above sensor noise.

In-band radiance is integrated from the Planck function over 8–14 µm for the source and the background. The fill fraction f that produces a k × NETD apparent rise is:

f = (L(Tb + k·NETD) − L(Tb)) / (M · τ(R) · (ε·L(Tf) − L(Tb)))

with M the optics factor and τ(R) = e^(−αR) for atmospheric extinction. Range follows from the pixel footprint, R = √(A / f) / IFOV, solved by bisection because τ depends on R.

This is the only figure that grade changes. Going from a 60 mK consumer core to a 20 mK Boson+ industrial core lowers the detection threshold, but the effect is modest next to resolution and focal length, because radiance rises steeply with source temperature.

Pixels on target

Defaults to the classic Johnson criteria: 1.5 cycles across the critical dimension to detect, 6 to recognise, 12 to identify, at two pixels per cycle. Critical dimension is the geometric mean of width and height. Grade has no effect here at all. Change the pixel counts in the Simulated view card below and this table, and the lens ruler further down, switch to your numbers too; a "custom" badge appears next to the title here when they do.

Angular resolution

Every Lepton and Boson in this tool uses a 12 µm detector, so range comes down to array size and focal length. Instantaneous field of view at frame centre is IFOV = pitch / EFL.

• Boson lenses have published effective focal lengths, so IFOV is taken straight from EFL. The manufacturer's stated horizontal FOV is shown alongside for reference; on the widest lenses (92° and 95°) the stated figure includes barrel distortion and runs wider than a pinhole model of the same EFL would give.
• Lepton 3.5 publishes only FOV. At 57° it is close enough to rectilinear that EFL is back-calculated from a pinhole model.
• Lepton 3.1R and UW are documented as barrel-distorted wide lenses at 95° and 160°, so they are modelled as equidistant (f θ), spreading angle evenly across the array. On the UW the extreme edges are heavily distorted and usable resolution near the border is worse than shown.

Bodies sold without a lens are omitted, since range cannot be computed until a lens is chosen. Frame rate options (60 Hz and 9 Hz) do not change range and are left out; they matter for fast-moving sources and for export classification.

The three Hadron 640-series modules (640R, 640+, 640R+) pair a Boson 640 thermal core with a 64 MP visible camera for image fusion; only the thermal side is modelled here, using the same 640 × 512 sensor and 13.6 mm / 32° lens already catalogued under Boson 640. GroupGets sells each model name as a single fixed-grade SKU rather than a Professional/Industrial choice, so the NETD used here (<40 mK for 640R's Boson core, <20 mK for 640+ and 640R+'s Boson+ core) is taken directly from each model's own GroupGets listing. Despite the "+" naming convention used elsewhere in this tool, GroupGets' own spec sheet lists the Hadron 640+ as non-radiometric, which is why only the "R" and "R+" variants are flagged radiometric here.

The simulated view

The frame is built at the module's real array size. Terrain is low-amplitude multi-octave noise around the background temperature, roughly half a degree of variation with a gentle warm gradient below the horizon. The target is drawn as an ellipse of the dimensions you entered and sampled onto the pixel grid; once it falls below about two pixels across it is treated as a point source and its flux is placed on the pixel it lands in, which is what produces the single bright blob.

Optics blur is a separable kernel whose centre weight equals the optics factor, so changing that control visibly softens the target. Sensor noise is Gaussian with a standard deviation equal to the selected grade's NETD. Display gain is a 0.5 to 99.5 percentile stretch, which is close to what a real core does and is why a fire clips to white while the terrain keeps its contrast.

The simulation includes terrain clutter, which the signature-limited calculation does not. That is deliberate, and the gap between them is the honest part. Push the range slider out and you will often see the target sink into the terrain before it sinks into the noise, which is exactly why the signature figure is a ceiling rather than a working range.

The "Pixels across target needed to" fields under the jump buttons default to the Johnson criteria used everywhere else in this tool (3 / 12 / 24 px, at 2 px per cycle). They drive the verdict shown under the frame, where the Detect / Recognise / Identify buttons jump to on the slider, the Pixels on target table, and the lens ruler above, all at once, so the whole tool reflects your own numbers once you change them. Only the hero signature-limited range stays put, since it is governed by NETD rather than a pixel count. Lower them if the thing actually reading the footage, whether a person or a model, can work with less than Johnson's criteria call for.

What this does not model

• Detector time constant and frame rate, which matter for fast-moving sources
• Automatic gain control, which can bury a small hot source in a high-dynamic-range scene. The Boson+ AGC is meaningfully better here than the numbers suggest
• Path radiance and solar loading on the background
• Flame spectral structure - open flames radiate in bands and are poor grey bodies in LWIR
• Lens f-number. All Boson lenses here are f/1.0 or f/1.1, so the difference is small