Airborne recognition

Recognition
that flies.

On the aircraft, Vektra is engineered to run beside the autopilot, not inside it. Detections, tracks and evidence are produced on board, shared with the flight stack over standard robotics interfaces, and held locally when the link is gone.

On the airframe

Perception that
keeps working.

The heaviest decisions about an aerial model are made before it flies: which module it runs on, how much power and heat it can spend, and what it must do when the operator’s feed goes dark. Vektra builds the model layer and the onboard pipeline together.

  • 01TensorRT engines compiled for the exact Jetson module that flies
  • 02EO and thermal streams in one hardware-accelerated pipeline
  • 03Detections published as ROS 2 topics and MAVLink messages
  • 04Evidence written on board before anything is transmitted
The aerial viewpoint

A model trained at street level
fails at fifty metres.

Most public imagery is taken from head height. Aerial recognition is its own domain, and the dataset has to be built for it.

Altitude

Objects shrink quickly as the aircraft climbs. Datasets span several survey altitudes, with scale augmentation in between.

Angle

Nadir, oblique and low-angle views. An object seen from overhead looks nothing like the same object seen from a doorway.

Bands

Paired visible and thermal imagery, with thermal-style augmentation of visible data where pairs are scarce.

Conditions

Motion blur, haze, rain, low light and partial occlusion, represented in training and in validation.

Look-alikes

Empty scenes and objects that resemble the target class, collected from the air on purpose.

Validation

A frozen aerial validation set, compared release by release on the target module.

Nadir inspection at three scales

The same site,
three scales down.

A substation from an inspection drone: the whole frame, one transformer bay, one radiator bank. Altitude and angle decide what a class looks like, so the dataset holds every scale the aircraft flies.

01 · Whole frame

Access lane, control building, three transformer bays and the switchyard beyond, from the inspection altitude.

02 · Transformer bay

One transformer with its radiator banks, conservator and bushings: the class an inspection model is trained to find.

03 · Radiator bank

Component scale: radiator fins, conservator tank and the cable trench beside them. Condition review starts here.

Integration with the flight stack

Speaks the aircraft’s
language.

Robotics middlewareROS 2 topics for detections, tracks and poses, with per-vehicle namespaces for multi-aircraft operation
AutopilotMAVLink messaging with PX4 and ArduPilot. Perception informs the mission; it never replaces the flight controller
GimbalTrack-driven cueing keeps an observation in view for the operator
VideoHardware-encoded RTSP streams with detection metadata alongside
GeometryAltitude, gimbal angle and range check that a detection’s size is plausible, and place observations on the map
Ground stationSecond-stage classification on ground compute and an operator review queue
Two stages. One decision.

Detect in the air.
Confirm on the ground.

01 / ONBOARD

Detect and track.

Dual-stream detection and multi-object tracking on the aircraft, so observations exist even when the link does not.

02 / GROUND

Classify and associate.

Larger second-stage models refine the class, re-associate tracks across views and rank candidates for review.

03 / OPERATOR

Confirm and record.

A person confirms or dismisses. Confirmation creates the evidence record, with its hash, time and position.

Export controls. Defense articles, technical data and defense services may be subject to the International Traffic in Arms Regulations (ITAR) and the Export Administration Regulations (EAR). We provide them only as authorized under U.S. export control laws, and we review end use and end user before any transaction. Do not send export-controlled technical data unless you are authorized to do so. Terms of use

Built around your challenge

What do you need
to recognize?

Start a project