AI Object Detection and Tracking: Detection and tracking of people and vehicles in live video, running on the aircraft rathe

Autonomy and AI[AI-TRACK]

Detection and tracking of people and vehicles in live video, running on the aircraft rather than in a data centre.

Runs on
Aircraft or field processor
Classes
People and vehicles
Connectivity
None required
Frame rate
On request

Overview

An operator watching a live feed for hours will miss things; that is a property of people, not of training. Detection that runs continuously does not get tired, and its job is to raise what is worth a human look.

The model runs on the aircraft or on a small processor beside the operator, so nothing depends on a link back to a server. Detections carry a class, a confidence and a track identity that persists as the subject moves, and can drive the sensor to keep a subject centred.

System qualities

01 / 04

On the aircraft, not in a cloud

Inference happens where the camera is. There is no upload, no latency penalty and no dependence on a link that may not exist.

02 / 04

Tracks, not just detects

A subject keeps its identity across frames, through partial occlusion, so a count is a count rather than a repeated detection.

03 / 04

Drives the sensor

A selected track can command the gimbal, keeping the subject centred while the aircraft flies its own task.

04 / 04

Raises, does not decide

Output is a flagged detection for a person to judge. The system does not take an action on its own.

  • AI Object Detection and Tracking, studio view
  • AI Object Detection and Tracking deployed in the field
  • AI Object Detection and Tracking, close detail

Specifications

Figures marked undisclosed or on request are released against a stated requirement.

Deployment

Location
Airborne processor or field workstation
Connectivity
Not required
Processor
Runs on the aircraft or on a field workstation

Detection

Classes
People, vehicles; extensible on request
Frame rate
On request
Sensors
Day and thermal video
Latency
On request
Minimum target size
On request

Tracking

Identity
Persistent across frames and occlusion
Sensor control
Selected track drives the gimbal
Simultaneous tracks
On request

Output

Interfaces
Overlay video, structured detections to the control station

Documentation

  • Autonomy and AI overviewAvailable on request

Find, follow, flag.

Next step

Brief us on your requirement.

Tell us the task, the terrain and the timeline. We will come back with what we would field and what it would take.