AI Thermal Counter-UAS

See drones first

// LIVE THERMAL · AI DETECTION & TRACKING

Passive detection and identification of unauthorized drones — built on AI vision, AI Edge devices and thermal sensors. No radar license. No spectrum emissions. No kinetics. Mobile-ready, deployable in hours.

Request a brief → See the system
ITAR-free · export-ready SAPIENT · BSI Flex 335 v2 C2 IP67 · sealed edge unit Passive · no spectrum license Made in Israel
// O-Lifeguard · at a glance
Passive AI thermal counter-UAS
≤400 m
Detection range
<20 ms
Edge AI latency
EO / IR
Layered passive sensors
On-edge
NVIDIA Jetson · offline-capable
SAPIENT
BSI Flex 335 v2 C2 output
IP67
−20 °C to +55 °C
Request Technical Data Sheet →
20+ yrs
IR sensor engineering
0
Emissions · fully passive
Day / Night
Identical performance
Hours
Field-deployable
+217%
// rising incursions
Rise in unauthorized drone activity across critical infrastructure versus pre-2020 levels — with European airports up as much as 300%.1
< $1k
// adversary cost
Mass-produced FPV attack drones are built from commercial parts for under $1,000 — cost ratios against the assets they target exceed 1,000:1. Asymmetric by default.2
< 1 m²
// below radar
Small drones sit in the radar-cross-section "grey zone" below the ~1 m² target legacy air-defense radar was built to track — too small, too slow, too low.3
seconds
// time to act
In a multi-drone scenario the defender has only seconds between critical decisions — automated detection and classification is essential.4

Sources: 1 Dedrone, Airspace Security Insights · 2 Global Security Review · 3 Robin Radar, Counter-Drone Technologies · 4 Lockheed Martin, C-UAS Challenge

// layered detection

Go dark. Go quiet. Still warm.

RF, acoustic and GPS denial are proven counter-UAS layers — but each leans on a signature or dependency a capable adversary can engineer away. GOTEAM's thermal layer covers exactly those gaps and fuses with the sensors you already run: the drone that slips past one still lights up on another.

Partner layer
// radio-frequency

RF detection

Reads the command-and-control link — long reach, strong against comms-heavy drones. Blind spot: a fiber-optic-tethered drone emits nothing on the spectrum. GOTEAM picks it up and cues it straight back into your RF picture.
Partner layer
// acoustic

Acoustic detection

Catches rotor noise and gives cheap bearing-to-target. Blind spot: low-noise "silent" propellers fall below the ambient floor at standoff range. Thermal cross-cues the array the instant a heat source appears.
Partner layer
// gnss denial

GPS jamming

Denies the drone its satellite fix to force a land or return-to-home — strong against GPS-dependent autopilots. Blind spot: fiber-optic, inertial and visual-navigation drones ignore GNSS entirely and fly straight through. Thermal sees the aircraft no matter how it steers.
Evasion-proof
// infrared

Thermal signature

Motors, ESCs and battery packs radiate heat the instant they spin up — thermodynamics, not a switch they can flip. Radio-silent and whisper-quiet, sure — but never cold. Run it stand-alone, or fuse it with RF and acoustic for a picture with no blind spot.
▶ See it detected · FlyBox →
// better together Every layer cross-cues the others through an open API — more detections, fewer false alarms, and higher operator confidence than any single sensor on its own. GOTEAM adds the one layer that can't be evaded, without replacing what you already trust.
// interoperability SAPIENT / BSI Flex 335 v2-compliant C2 output plugs straight into your existing command-and-control. GOTEAM cues azimuth / elevation / range to third-party effectors and fuses with the RF and acoustic layers already in the picture — an open, standards-based partner layer, not a closed silo.
// the system

Detect → Engage

A passive optical sensor stack with on-edge AI classification. No spectrum emissions. No moving parts. Single operator, single screen.

// optical detection

EO/IR vision stack

Layered electro-optical and infrared sensors. Daylight, low-light, and thermal. Passive — your installation never broadcasts. The adversary doesn't know they've been seen.
▶ Run it on your own clip · FlyBox → See for yourself · FlyOps →
// EO/IR · TRACK CLASS-1 · DRONE RNG · 13.84m CONF · 0.83
// field-ready · IP67

Field Ready AI

Neural classification of drone vs. bird vs. clutter, plus class identification across 100+ commercial and military drone signatures. Inference runs at the sensor on a fanless, IP67 ruggedized NVIDIA edge compute box — dust-tight and immersion rated, no cloud round-trip, field-deployable, vehicle-grade DC power.
// EDGE COMPUTE NVIDIA · FANLESS · IP67 CUDA · ON-DEVICE DC 9–36V · FIELD-DEPLOYED
// tracking

Multi-target tracking

Persistent tracks across the field of view. Trajectory prediction, threat scoring by behavior, and prioritization for the operator. Hand-off to other systems via open APIs.
// form factor

Mobile + fixed

Vehicle-mounted, tripod, or rooftop. Power: vehicle DC, mains, or solar+battery. Deploy in hours, relocate in minutes. No civils, no spectrum permits, no kinetic licensing.
// solutions

Where it goes to work.

Built for environments where the threat moves faster than legacy radar — and where emitters or kinetics are not an option. Two deployment modes, one passive sensor.

// mobile · tactical
Thermal night operations — Armor and convoys
// 01

Armor & convoys

Tanks, armored vehicles and moving convoys. Vehicle-mounted passive detection that travels with the column — no emissions to give away the route.
Thermal night operations — Bases and field camps
// 02

Bases & field camps

Forward operating bases and field camps. Standalone or as a sensor layer feeding existing C2 — deployable in hours, relocatable in minutes.
Thermal night operations — VIP protection
// 03

VIP protection

Motorcades, residences, off-site engagements — worldwide. Mobile detection that travels with the principal; passive, discrete, hand-offable to security teams.
Thermal night operations — Police and events
// 04

Police & events

Crowd events, prison perimeters, urban deployments. Civilian-airspace appropriate, no jamming side-effects, full audit log.
// fixed · strategic sites
Thermal night operations — Airports
// 05

Airports

Runways, approach corridors and terminal perimeters. Continuous passive awareness without adding to a crowded RF and radar picture.
Thermal night operations — Critical infrastructure
// 06

Critical infrastructure

Energy — power, grid, oil & gas — plus water, ports and dams. Continuous perimeter awareness without disrupting the local RF noise floor.
Thermal night operations — Defense sites
// 07

Defense sites

Defense companies, production plants and hardened installations. Standalone or fused into an existing multi-sensor C2 layer.
Thermal night operations — Government and symbolic sites
// 08

Government & symbolic sites

Parliaments, ministries and symbolic landmarks. Passive, spectrum-clean detection appropriate for dense civilian airspace, with a full audit log.
// technology

Sensor → Model → Detection.

Three stages, all on-edge. The pipeline runs locally on the deployed unit; no internet dependency in operation.

// stage 01
Optical capture
EO + IR sensors
full-frame at 30+ FPS
configurable PTZ
weather-rated
→
// stage 02
AI inference
classification + tracking
100+ drone signatures
on-edge (no cloud)
continuous learning
→
// stage 03
Decision layer
threat scoring
operator screen
API hand-off
full audit log
no spectrum emissions · no kinetic action · open APIs to existing C2
// edge compute
On-device AI. No cloud.
A sealed IP67 edge-compute box runs every stage of detection on-device on NVIDIA Jetson — classification, tracking and threat scoring all happen at the sensor. No cloud dependency: there is no round-trip to a server and nothing to intercept in flight. What the sensor sees, the sensor decides.
// deploy anywhere
Runs offline. Backhauls when it can.
The unit is fully offline-capable — it needs power and a field of view, nothing else. When a link is available, it backhauls detections and imagery over whatever the site has: Ethernet, cellular, or Starlink.
Offline Ethernet Cellular Starlink
Passive
EO/IR
// sensing
up to
400 m
// detection range
<20 ms
// edge latency
IP67
// sealed rating
SAPIENT
Flex 335 v2
// c2 output
ITAR-free
// export
// how it works

How AI thermal drone detection works

The hard part was never the camera. It's pulling a few warm pixels of drone out of a live sky full of cloud, birds and sun-heated clutter — reliably, at low false-alarm rates, on a device in the field. Here is the path from a raw thermal frame to a confirmed, ranged track. Every stage runs on the edge unit.

  1. Raw thermal capture

    The uncooled thermal core streams full 16-bit frames at the sensor's native bit-depth — not a compressed 8-bit picture. That raw depth is what lets a faint target survive to the next stage.

  2. Suppress the sky

    A morphological stage strips large, slow-changing structure — clouds, sky gradients, warm horizon — and leaves only the small, sharp hot peaks. A drone against a bright cloud, invisible to the eye, pops out cleanly.

  3. Neural detection

    A purpose-trained neural network — built on tens of thousands of hand-verified thermal drone frames — scans every processed frame and proposes targets, down to a handful of pixels across.

  4. Confirm, don't flicker

    A tracker links detections frame-to-frame by position and velocity, and only confirms a target once it persists and moves consistently. Single-frame noise and static hot spots are gated out — that's what keeps false alarms low.

  5. Range & hand off

    Each confirmed track gets a calibrated distance, azimuth and elevation, then streams to the GOTEAM hub as a live map track with an annotated clip — and out to any connected C2 over SAPIENT.

A moving PTZ camera watches one sector at a time and needs a radar cue to know where to look. A fixed GOTEAM node images the whole scene every frame and searches it with AI — so a single drone or a whole swarm is detected the same way, with no blind time and no cue required.

Is uncooled LWIR thermal good enough to detect small drones?

Yes — for perimeter and near-airspace defence, which is where the small-UAS threat lives. Uncooled long-wave infrared reads the contrast between a drone's airframe and a cool sky very well at those ranges. The reach that used to require an expensive cooled sensor now comes from the detection software: GOTEAM trains its models on real field data so an affordable passive core delivers confident detections without a cryocooler, a spectrum licence or any emissions.

LWIR vs cooled MWIR for counter-UAS — what's the difference?

Cooled mid-wave (MWIR) sensors are more sensitive and see farther, but cost ten to fifty times more, draw far more power, carry a cryocooler with a limited service life, and usually come with export-control restrictions. Uncooled long-wave (LWIR) is inexpensive, low-power, rugged and emission-free. GOTEAM pairs LWIR with AI to get high recall inside the envelope where you can actually act on a small drone — instead of paying for range you can't engage.

Can it detect autonomous or radio-silent drones?

Yes. GOTEAM detects the aircraft itself — its shape and heat — not its radio link, so a fibre-tethered, pre-programmed or fully autonomous drone that transmits nothing is just as visible. Detection is passive: the sensor emits nothing and cannot be jammed or geolocated by the target.

How does it handle multiple drones or a swarm at once?

Because the sensor is fixed and stares at the whole scene, the AI runs detection across every target in the field on every frame — one drone or twenty, at the same latency. There is no gimbal that has to choose which target to point at, so track custody isn't lost as the count climbs. Multi-target is native to the architecture, not a scheduling problem.

// live sandbox

Try it on your own footage.

FlyBox is a public sandbox of the detection pipeline. Upload a thermal or EO clip and a GOTEAM Field device processes it on real hardware — same models, same inference path as a deployment. The annotated result streams back to your browser.

Upload a clip. See the system run.
Drop a video or a zip of frames (up to 500 MB). Add an optional flight-log file for synchronised metadata. We queue it on a Field device, run drone-detection locally, and stream the annotated MP4 back — usually in minutes. No install. No integration. Up to 17 jobs per email for evaluation; talk to us for more.
Thermal & EO 500 MB / clip Field-device inference Annotated MP4 No install
Open FlyBox →

More than a sandbox, FlyBox is the command and monitoring layer for every GOTEAM sensor in the field — a cloud hub that ties your deployed nodes into one live picture, on the desk and in your pocket.

Live node map
Every deployed node on one map — status, health and coverage at a glance, across sites and vehicles.
Live thermal feeds
Stream the annotated thermal feed from any node in real time — the same AI detections the operator sees in the field.
Detection review
Replay and triage every detection — confirm, dismiss and export events with imagery and metadata attached.
SAPIENT map
A live SAPIENT / BSI Flex 335 v2 tactical picture — detections from GOTEAM and partner sensors plotted on one shared map.
Mobile app
Native iOS app and an installable PWA — the whole hub, on a phone, wherever the mission is.
One picture, many nodes
Store-and-forward keeps offline nodes in sync; when a link returns, their events and imagery roll up into the same hub view.
// operational scenarios

The system, scenario by scenario.

How GOTEAM handles real operating conditions. Each clip is live model output — the boxes are produced on a Field device, on thermal footage captured in the field. No staging, no post-editing of the detections.

▶ Try it on your own clip · FlyBox →
Up · Cloud · Sky

Drone against open sky

A target climbing and then hovering against broken cloud. The tracker holds the drone through the hover — the hardest moment, when a near-stationary target is easiest to lose.
Open video
Final Flight

Final Flight

The closing seconds of a tracked engagement, looped — the system staying locked on the target right through the end of the run.
Open video
Walk in the Woods

Walk in the woods

A target tracked over a wooded, broken-terrain background — thermal clutter from foliage and ground that the system holds the lock through.
Open video
// questions

RF, acoustic, optical — where each one fits.

The questions buyers who already run counter-UAS sensors ask us most — how passive optical detection works, and how it covers the blind spots in RF and acoustic.

01How does GOTEAM detect drones without radar or RF?
GOTEAM is a passive AI-optical counter-UAS system. It detects and identifies drones from their visual and thermal signature using EO/IR cameras and on-edge machine learning — no radar, no transmitter, no spectrum license. Because it emits nothing, it can't be jammed or geolocated by the threat.
02Can it detect drones with no RF signal, like fiber-optic-controlled drones?
Yes. RF detection finds a drone by listening for its command-and-control radio link, which makes it blind to a fiber-optic-tethered or fully autonomous drone that transmits nothing. GOTEAM detects the aircraft itself — its shape and heat — so a radio-silent drone is just as visible. Where you already run RF sensors, GOTEAM cross-cues them: it flags the target your spectrum sensor can't hear and hands off a bearing.
03What about quiet or "silent-propeller" drones that defeat acoustic sensors?
Acoustic arrays key on rotor and propeller noise and give low-cost bearing-to-target, but low-noise silent propellers drop below the ambient noise floor at standoff range. A drone's motors, ESCs and battery still radiate heat the instant they spin up — thermodynamics, not a setting — so GOTEAM sees it even when the microphones go quiet. The two layers cover each other.
04Does GOTEAM replace our existing RF and acoustic counter-UAS sensors?
No — it complements them. RF, acoustic, radar, GPS-denial and optical layers each have blind spots. GOTEAM adds the evasion-proof thermal and optical layer and fuses with the others through an open API. Every layer cross-cues the rest: more detections, fewer false alarms, and higher operator confidence than any single sensor alone.
05Does GOTEAM jam, intercept, or shoot down drones?
No. GOTEAM is detection and identification only — passive, no kinetics, no jamming. It gives operators early, accurate situational awareness and feeds your existing decision and mitigation layer.
06What does GOTEAM need to deploy?
Power and a field of view. GOTEAM is mobile-ready and fixed-site capable, needs no spectrum license and no emissions authority, and deploys in hours. On-edge inference means it runs without network backhaul.

Want a technical brief?

Not a generic deck. A specifications-first conversation tailored to your mission profile, threat envelope, and integration constraints. Government, military, and qualified end-users only.

// vetted inquiries only · response within 2 business days · NDA-ready