What hardware does the ATR run on?

The Neuronic Video Analytics Suite is an embedded stack that runs on NVIDIA Jetson class edge hardware onboard the platform, so no ground processing centre is required.

How does ATR output reach our command system?

Target metadata is published as STANAG 4609 / VMTI, JSON, ROS or MAVLink, and can be cross-cued into NOMAD or ATAK.

What is the difference between ATR and an AOI tracker?

ATR automatically finds and classifies targets in a video feed. An AOI tracker keeps a specific target or area in view and centred, often driving the gimbal or cueing other systems.

Can it run on my existing drone or observation post?

Yes. The suite runs on NVIDIA Jetson class computers and accepts common video inputs, so it can be added to existing UAS, UGVs, observation posts and gimbal payloads.

Does it need a cloud or ground processing station?

No. All processing happens at the edge, so the capability survives datalink loss and contested networks.

Perception: detect, track and locate targets at the edge

Edge AI turns EO and IR video from drones and ground platforms into target information.

Automatic target recognition overlay tracking targets in a live EO video feed

Neuronic Perception runs detection, classification and multi-target tracking onboard the platform, and outputs tracks and STANAG 4609 metadata to your existing systems.

Products

Video Analytics Suite

Perception & targeting

ATR Detector-Tracker & AOI Tracking

Automatic target recognition (ATR) is software that finds, classifies and tracks objects in live EO or IR video. Neuronic runs ATR on the platform itself, so detections, persistent track IDs and AOI tracking are produced onboard and published as STANAG 4609 / VMTI, JSON, ROS or MAVLink metadata rather than streaming raw video to a ground station.

Step 1

Ingest live EO or IR video

Video is received from gimbals, observation posts, UAV payloads or STANAG 4609 streams over RTSP, UDP/RTP and direct camera inputs.

Step 2

Detect and classify

Edge models locate objects of interest, assign class labels and give each target a persistent identity that survives crossing and grouping.

Step 3

Track and re-acquire

Multi-target tracking holds IDs through brief occlusion, and AOI tracking keeps a selected target centred for gimbal or effector cueing.

Step 4

Publish target metadata

Structured target information flows to C2, autonomy and gimbal control over STANAG 4609 / VMTI, JSON, ROS or MAVLink.

The operator cannot watch every pixel

Modern sensors generate more video than operators can monitor. Missing a target in a feed, losing track during occlusion, or waiting for a datalink to process the video all slow the detect-to-decision cycle. ATR solves this by processing video onboard and reporting only the targets that matter.

Why it matters

Reduces operator workload by turning video into a track list.
Keeps working when bandwidth is jammed or denied because processing is at the edge.
Supports day and night operations through EO and IR pipelines.
Enables counter-UAS detection and track hand-off to effectors or command systems.

Demos

See it working

ATR Real-Time Detection & Tracking

Neuronic edge analytics detect, classify and track multiple targets in real time from EO and IR video, holding track identity through occlusion and platform motion.

All demos

Quick answers

Questions about Perception

What hardware does the ATR run on?: The Neuronic Video Analytics Suite is an embedded stack that runs on NVIDIA Jetson class edge hardware onboard the platform, so no ground processing centre is required.
How does ATR output reach our command system?: Target metadata is published as STANAG 4609 / VMTI, JSON, ROS or MAVLink, and can be cross-cued into NOMAD or ATAK.
What is the difference between ATR and an AOI tracker?: ATR automatically finds and classifies targets in a video feed. An AOI tracker keeps a specific target or area in view and centred, often driving the gimbal or cueing other systems.
Can it run on my existing drone or observation post?: Yes. The suite runs on NVIDIA Jetson class computers and accepts common video inputs, so it can be added to existing UAS, UGVs, observation posts and gimbal payloads.
Does it need a cloud or ground processing station?: No. All processing happens at the edge, so the capability survives datalink loss and contested networks.

Other technologies

Monitored Autonomy Tactical Hybrid Swarm Control

Discuss your platform and mission

Tell us the platform, sensor and mission. We will walk you through integration options and arrange a technical demonstration.