DG Defence // Perception Systems
Eyes that name
the threat.
AEGIS Vision is an open-vocabulary perception engine for autonomous flight. It detects, segments and tracks anything your mission names, including the targets closed-vocabulary detectors are blind to.
A drone is only as capable as
what it can recognise.
Autonomous flight puts the perception burden on the aircraft. In contested airspace, the payload must detect, classify and track threats it has never been explicitly trained to name, in real time, at the edge.
See everything
A drone's sensors capture the whole scene, but raw pixels aren't perception. The payload has to know what it's looking at, instantly, on-board.
Name the unknown
Adversaries field novel platforms: humanoid units, unfamiliar drones, improvised systems. A detector trained on a fixed list simply doesn't have a word for them.
Hold the track
Counting a target once isn't enough. Perception must lock identity across frames, through occlusion and motion, to feed targeting and BDA downstream.
Same frame. Two worlds.
A live-fire sortie, both models run frame-for-frame. Drag the handle to reveal what the baseline reports versus what AEGIS Vision segments and tracks.
AEGIS Vision
BASELINE · OthersHumanoid units advancing under arms. Baseline recognised 0; AEGIS Vision segmented 14.
AEGIS Vision holds a 127% steadier confidence signal than the baseline on this clip.
Measured, not claimed.
Every figure below is computed frame-by-frame from the same footage the demos run on: detection rate, confidence and per-frame stability across three clips.
AEGIS Vision tracks the full formation continuously; the baseline collapses toward zero whenever a unit turns, occludes or leaves its known distribution.
One eye. Every platform.
AEGIS Vision is a perception core, not an airframe. The same eye that flies a quadcopter bolts to a fixed-wing, a patrol vehicle, an armoured hull or a quadruped. Power and a data line, and it sees.
One engine, open-ended.
AEGIS Vision replaces the fixed class list with language. Describe the target and it finds it, then segments, tracks and holds it, on the airframe.
Open-vocabulary prompting
Prompt it in plain language. AEGIS Vision localises whatever you name, with no retraining and no fixed class list. New threat profile at 08:00? Type it and fly.
Segmentation, not boxes
Pixel-precise masks give exact footprint, pose and extent for every object: the geometry targeting and BDA actually need.
Persistent tracking
Every object keeps a stable ID through occlusion, motion and re-entry, so counts and tracks don't reset frame to frame.
Edge-deployable
Engineered to run on the airframe: bounded latency, fully offline, no reach-back to a ground station required.
Field readout · live-fire range
Every unit. Every frame.
Named and held.
Start the conversation.
We work with defence primes, integrators and national programmes. Tell us the mission and we will arrange a technical walkthrough of AEGIS Vision against your own footage.
Write to us directly
One click opens a pre-filled inquiry. Add your company details and send.
Opens Gmail with subject and message ready. Prefer your own client? netan@disobedientgeometry.com
Handled under NDA. No data leaves your environment during evaluation.