Drone detection for critical infrastructure

Small non-cooperative drones are cheap, quiet and easy to obtain. For industrial, energy, research and other sensitive sites this creates several effects that require no weapon payload at all:

• Surveillance. Plant layout, prototypes, construction progress, fence lines, access roads and shift patterns can all be documented from the air.
• Operational disruption. Even a well-founded suspicion of a drone can halt work, clear areas or delay processes.
• Preparatory activity. An overflight may precede a later act on the ground.
• Unclear responsibility. For many security organisations it is not settled who may do what when a drone is sighted.

What these cases have in common: the operator usually never learns that anything happened at all. Without detection there is no sighting, without a sighting no documentation, and without documentation no defensible risk assessment.

At the detection layer there are essentially three sensing principles, and they differ in their blind spots:

Radar is therefore not a replacement for the other methods but closes their largest gap: it needs neither a radio signal nor visible light, and it simultaneously supplies the positional information a camera needs before it can be pointed usefully. The price is that radar alone does not say what is flying – which is exactly what Ozense works on.

Detection and alert assessment are intended to run entirely on site, on one edge unit per radar node. For sensitive sites this has several practical consequences:

• No cloud dependency for detection. Availability does not hinge on an internet connection.
• No raw-data transfer across the site network in the current architecture.
• Data minimisation becomes easier. The radar channel produces no images; a camera is only foreseen event-driven, with no continuous video recording.
• Lower latency between detection and alert.

None of this amounts to blanket GDPR compliance. Radar data can become personally identifiable in combination with other information, and legal basis, field of view, retention period, access and deletion remain to be assessed per installation. Local, event-driven processing makes that assessment easier – it does not replace it.

Radar produces candidates, the tracker turns them into temporally stable tracks, the validity stage discards clutter and artefacts, and only then is a valid track assessed semantically. Where evidence is insufficient, the result is Unknown. Only for a relevant candidate should an existing PTZ camera be slewed, and only where visibility and target geometry allow.

Each stage reduces the alert volume reaching the next. This chain is not described by a single model figure but measured stage by stage – from the raw candidate through to what actually appears in the control room. Details on the processing chain.

An alert is only worth something if somebody receives it and is permitted to act on it. The target architecture therefore hands events to existing security technology rather than building a parallel island: a time-stamped incident package comprising the track, any image material, confidence, and configuration and version state, delivered to a VMS, hazard management system or control room.

The operator alone decides on the response: observe, document, exercise property rights, report, or hand over to the responsible authorities. Ozense performs no neutralisation and confers no power of intervention.

No particular VMS vendor and no list of compatible PTZ models have been fixed so far; productive integration is a subject of the funding phase.

The German KRITIS Umbrella Act has been in force since 17 March 2026. It names environmental monitoring and detection equipment among possible resilience measures and allows evidence to be requested on a risk basis. Registration and deadlines depend on how the individual facility is classified, so no single date applies to all operators.

What does not follow: a blanket statutory obligation to procure a drone radar. The need must be justified per site from risk analysis and proportionality. Regulation can raise the pressure for evidence but does not replace an assessment of the specific case.

For operation itself: the general assignment covering automotive short-range radar does not make fixed-site operation automatically lawful. Frequency clearance is handled separately for own research, for fixed pilot operation on third-party premises, and for later product operation; an application for research use is pending with the German Federal Network Agency.

A detection system whose performance nobody can verify is hard for a security organisation to assess. Ozense is therefore developing an evidence module under the working name FlightAssure: pre-registered function and acceptance tests, ground-truth-bound evaluation with confidence intervals, defined Unknown and degradation paths, data, model, configuration and software provenance, and repeatable checks after an update, maintenance or site change. The EU pre-standard CWA 18150 / COURAGEOUS serves as methodological orientation.

FlightAssure replaces neither a statutory certification nor a site-specific compliance assessment. Nor does it guarantee a field false-alarm rate; it makes the performance actually measured, and its limits, traceable.