Ground control station tablet and laptop showing ATR bounding boxes tracking a drone in a live video feed
    Back to all solutions

    Detector & Tracker

    ATR

    Automatic Target Recognition for carts, tablets and airborne platforms

    Download ATR datasheet (PDF)

    A single detection and tracking engine that runs identically on the ground control cart, the operator's rugged tablet and the aircraft itself. Operators keep the same target picture whether processing happens at the station or onboard, so hand-off between ground and air never resets the track.

    Detection

    • Multi-class ATR for personnel, light and heavy vehicles, vessels and static structures
    • EO/IR and multispectral input with per-sensor model profiles
    • Small-object detection tuned for high-altitude and long-standoff imagery
    • Confidence gating and operator-adjustable classification thresholds

    Tracking

    • Persistent multi-object tracking with re-acquisition after occlusion
    • Operator click-to-track and auto-lock on highest-priority classification
    • Track continuity across zoom, gimbal slew and platform manoeuvre
    • Geo-registered track output for fusion into the common operating picture

    Deployment surfaces

    • Ground control carts — rack or box compute driving multi-feed operator displays
    • Rugged tablets — dismounted operation with the same track pipeline at reduced power
    • Onboard drone integration — direct feed from payload to mission computer, no downlink required

    Integration

    • ROS2, MAVLink and custom protocol bridges
    • Track and detection streams over standard video and telemetry channels
    • Deterministic behavior with full audit trail of detections and operator actions
    • Runs disconnected — no cloud dependency in contested comms

    At a glance

    Compute
    NVIDIA Jetson (Orin family), Qualcomm
    Sensors
    EO, IR, multispectral
    Form factors
    Cart, tablet, airborne payload
    Interfaces
    ROS2, MAVLink, custom
    Connectivity
    Fully offline capable
    NVIDIA JetsonQualcommiMX8Hailo

    Integration FAQ

    Which compute platforms does the ATR detector and tracker run on?
    It runs on NVIDIA Jetson Orin modules and Qualcomm compute, with additional builds for iMX8 and Hailo accelerators. The same model pipeline is used across ground control carts, rugged tablets and onboard drone computers, so behavior stays consistent when a mission moves between form factors.
    How is ATR integrated into an existing ground control station or airframe?
    Integration is through ROS2, MAVLink or a custom protocol bridge. Detections and tracks are published over standard video and telemetry channels, so an existing operator display can overlay them without replacing the ground control software.
    Does ATR require a network or cloud connection?
    No. All detection and tracking runs at the edge on the host compute. The system is fully offline capable and continues to operate in contested or comms-denied environments with no cloud dependency.
    Which sensors are supported?
    EO, IR and multispectral payloads are supported, with per-sensor model profiles. Small-object detection is tuned for high-altitude and long-standoff imagery, and tracks persist across zoom, gimbal slew and platform manoeuvre.
    How long does a typical integration take?
    Timelines depend on compute, sensor set and platform interfaces. Share those three details with our team and we will scope the integration, including any model tuning needed for your imagery.

    Integrate ATR into your platform

    Tell us your compute, sensors and platform, and we will scope the integration.