Home

— Wildfire detection · Edge AI · Real Time

Autonomous wildfire detection.
Embedded. Real-time.
No cloud dependency.

NVIDIA Jetson Orin NX · C++ / DeepStream · L1 → L2 → L3 Kill Chain

NumidAI turns a single edge device and any RTSP camera into an autonomous wildfire monitoring system — designed to detect weak smoke signals and verify them locally before issuing an alert.

The system uses a staged on-device verification pipeline: a first detection layer scans continuously, deeper analysis stages re-check persistent candidates, and only consistent signals are escalated.

Built for real-world outdoor conditions, NumidAI is designed to separate credible wildfire smoke from common visual confusers while running fully on-device.

Technical demonstrator. Open to discussions around pilot validation in real-world conditions from Summer 2026 onward.


Demo

Two real-world cases: prescribed burn detection · industrial plume rejection


Kill Chain

Kill chain: Camera → L1 Sentinelle → L2 Sniper → L3 Semantic → ALERT
Each stage eliminates candidates. Only signals surviving all three verification levels produce an alert.

L1 — Sentinelle

Wide scan · Weak-signal detection

Wide-area continuous detection, built to pick up weak and uncertain signals early.

L2 — Sniper

HD crop · Dynamic and temporal verification

Persistent candidates are re-checked locally through higher-fidelity analysis, with dynamic and temporal consistency checks used to distinguish stable smoke signals from transient artifacts before escalation.

L3 — Semantic

On-device source classification · Ambiguity resolution

Ambiguous survivors go through a final on-device semantic verification stage, designed to identify known non-threat sources and resolve the remaining uncertainty before alert escalation.


Stack

Hardware

  • NVIDIA Jetson Orin NX 16 GB
  • Any IP camera · RTSP H.264/H.265
  • No cloud compute required

Software

  • C++20 · NVIDIA DeepStream 7.x
  • YOLOv8m · TensorRT FP16
  • Embedded semantic model
  • Lock-free async architecture

Contact

For field validation partnerships, technical collaboration, or any inquiry — see the About page →