The camera that can only look one way

Walk any counter-UAS trade floor and you will be sold on range. Cooled mid-wave sensor, continuous zoom, "detection out to five kilometres." It is an impressive spec sheet. It is also, for the threat most sites actually face, the wrong thing to optimise — and the architecture it forces on you carries risks the brochure never lists.

As an engineer I want to separate two questions people keep merging: which thermal band you use, and whether the camera moves. The first is a sensor choice. The second decides whether you can see a swarm at all.

Two kinds of thermal, honestly

Thermal counter-UAS sensors divide by the infrared band they capture:

• Uncooled LWIR (long-wave, 8–14 µm) uses a microbolometer — no cryocooler, low power, mechanically simple, inexpensive and rugged. It reads the contrast between a drone's airframe and a cool sky very well at the ranges that matter for perimeter defence.
• Cooled MWIR (mid-wave, 3–5 µm) uses a photon detector chilled to cryogenic temperature by an integrated cooler. It is genuinely more sensitive and reaches farther. It also costs ten to fifty times more, draws far more power, needs cool-down time before it works, ships with a cooler whose service life is measured in hours, and usually arrives with export-control baggage.

Both are real tools. The honest question is never "which sensor is better" — it is "which architecture fits the mission." And the high-end MWIR-plus-gimbal tier was inherited from platforms built to watch for large, fast aircraft. Point it at the small-drone problem and the seams show.

The scanning gap

A pan-tilt-zoom sensor sees one narrow field of view at a time. To cover a perimeter it slews or step-stares from sector to sector — which means that while it is looking at sector A, a drone crossing sector B is simply not observed. There is always blind time, and an adversary only has to arrive during it.

High-end systems paper over this by not searching at all — they wait to be cued by radar or an RF analyser, then slew to the bearing. That works beautifully against a target that announces itself. It fails against exactly the threat that is growing fastest: a small, autonomous, radio-silent quadcopter with a radar cross-section barely above the noise. No cue, no slew, no look, no detection. The most dangerous drone is the one your camera was never told to turn toward.

One camera, one target

Here is the question that ends the debate: how do you track multiple targets with a PTZ?

You can't — not really. One aperture points one direction. To follow several targets it has to time-slice: slew to A, dwell, slew to B, dwell, and during every dwell every other target goes unwatched. Track custody degrades as the count climbs; tracks drop, get mis-associated on re-acquisition, and the picture falls apart precisely when it is most crowded. A coordinated swarm is not an edge case for a trainable sensor — it is the designed-in failure mode.

Fixed-stare inverts this. Every target in the field is imaged in every frame, and the detector runs across the whole frame on every pass. One drone or twenty, the latency is the same and no track is ever put down to go look elsewhere. Multi-target is not a scheduling problem to be managed. It is just how the system already works.

Range is a vanity metric

"But we detect at five kilometres." Do you care? Two things collapse the long-range flex for the small-drone fight:

• Your effectors are short-range. RF jammers, nets, interceptors, most directed energy — their engagement envelopes are frequently well under a kilometre. A track at five klicks you cannot touch until five hundred metres is a number, not a defence.
• Identification needs pixels on target. A small drone at five kilometres is a warm smudge — a track, not an identification. You cannot confidently call drone-versus-bird until it is closer, on any sensor. The actionable decision happens inside the near envelope regardless of how far the core can technically reach.

So the real job is not seeing farther. It is confident, simultaneous, multi-target detection inside the envelope where you can actually act.

Where does long range legitimately matter? Fast fixed-wing threats, standoff attack, or critical infrastructure with long, open approaches. If that is your mission, buy the range. For the small-UAS perimeter threat — the one most sites are losing sleep over — it is the wrong optimisation.

How GOTEAM gets there instead

We put the intelligence in the software, not a half-million-dollar sensor. GOTEAM Flyer stares — wide field, no moving parts, no blind time, no cue required — and lets a purpose-trained neural network do the reaching:

• Capture the raw 16-bit thermal frame — full sensor bit-depth, not a compressed 8-bit picture, so the faint targets survive to the next stage.
• Suppress the sky and clutter, leaving only the sharp hot peaks a drone makes against it.
• Detect with a network trained on tens of thousands of verified thermal drone frames.
• Confirm each target across frames by motion — so sensor noise and static hot spots never raise an alarm.
• Range and hand off — a live map track and annotated clip to your hub and any connected C2.

And there is a quieter advantage that matters in the field: with no gimbal motors, no slip rings and no cryocooler, there are no wear items to fail. A fixed-stare node's mean time between failures (MTBF) isn't dragged down by a mechanism that has to move thousands of times an hour or a cooler with a rated service life — so it keeps watching, unattended, far longer between service visits.

No gimbal to fail. No sector it wasn't watching. No swarm it can only half-see. And affordable enough to blanket a perimeter instead of guarding it with one expensive eye that can only look one way.

If you protect a site, a perimeter or a fleet, we'll show you the system on real hardware, with real detections: goteam.co.il/management.html

— Baruch