Ethical Facial RecognitionSupported
FaceTrack Pro accepts mobile and drone sources. Its resolution requirements still bind at roughly 500 pixels per meter and 50 pixels eye to eye, so altitude and lens decide whether a match is possible.
Drone footage, searched like any other camera. A drone is a camera that moves, and once its footage reaches the platform it is searched, matched and audited exactly like any other source. There is one honest qualification: which modules apply depends on the angle you flew, not on the aircraft.
How It Works
Aerial video is ingested the same way any non-fixed source is: uploaded through Gallery › Uploads, then processed with Analyze video. The upload becomes a first-class channel with its own timestamp, location and access-control group, so the resulting events appear in the ordinary event filters beside events from fixed cameras.
That also means it inherits the platform's accountability rather than sitting outside it. Analyzing an upload is bound to a Case ID and written to the non-erasable audit log, the same as a watchlist search or an archive query. There is no separate, lighter-touch path for aerial footage.
What it does not mean is that every module applies. Suitability is decided per module by viewing geometry, and the honest table is below.
Per Module
FaceTrack Pro accepts mobile and drone sources. Its resolution requirements still bind at roughly 500 pixels per meter and 50 pixels eye to eye, so altitude and lens decide whether a match is possible.
Plate, make, model and color, where the viewing angle is about 10–20° below the horizon. That is the same geometry a mast-arm enforcement camera uses, and a steep top-down pass does not satisfy it.
A condition described in plain language becomes a detector on drone video as it does on a fixed stream. StreamVLM is shipping in beta on selected instances.
FireTrack, ObjectTrack Pro, MotionTrack Pro and CarCount are documented as fixed-camera only and explicitly exclude drone and mobile sources. Person re-identification runs on ObjectTrack Pro and inherits that limit.
A prompt-defined detector catches a drone launching or landing on an ordinary camera. That detector is already live on our own home page. What it cannot do is follow an aircraft through the air: StreamVLM scores sampled frames rather than running at the frame rate a track needs.
IREX supplies analytics, not flight operations. Airspace class, permissions and aviation law in your jurisdiction are the operator's responsibility, and nothing here implies a right to fly.
Why the Angle Matters
Vehicle recognition was built for a camera that looks along a carriageway from a modest height, roughly 10 to 20 degrees below the horizon, and sees plates near square to the lens. A drone that holds that geometry is, to the analytics, a mast arm that can move. A drone directly overhead is not: at 90° there is no plate face to read and no vehicle profile to classify, however good the sensor is. The practical consequence for a flight plan is that a shallow standoff pass returns results and a top-down survey pass does not.
License Plate Recognition →No. The documented workflow is file-based: upload the recorded footage through Gallery, run Analyze video, and search the resulting events with the ordinary filters. That is what makes it work with the drones an agency already flies rather than requiring a particular airframe or radio.
FireTrack, ObjectTrack Pro, MotionTrack Pro and CarCount are documented as fixed-camera only, and each explicitly excludes drone and mobile-phone sources. Person re-identification runs on ObjectTrack Pro and inherits the same limit. We would rather tell you that in a page than in a pilot.
Exactly the same. Recognition is restricted to pre-registered people, every search is bound to a Case ID, and every action is written to a non-erasable audit log. Aerial capture changes the camera, not the accountability.
Two different questions. A drone launching or landing is a discrete event on the ground, and a StreamVLM™ prompt on an ordinary camera detects it today. That detector is live on the IREX home page. Following an aircraft through the air is not possible: StreamVLM scores sampled frames rather than running at the frame rate a continuous track requires. Stopping one is counter-UAS, a sensor and effector problem rather than a video-analytics one, and we do not sell it as ours.
Keep Reading
The fastest way to settle whether your footage is usable is to analyze a real sortie rather than argue about altitude in a meeting.