GitHub Readme

Deterministic Threat Detection#

Welcome to the documentation for the Deterministic Threat Detection project—a Time-Sensitive Networking (TSN) demonstration showing how to deliver deterministic, low-latency AI and sensor workloads in shared networks.

Overview#

Deterministic Threat Detection Architecture

Component

Role

MOXA TSN Switch (TSN-G5000)

PTP Grandmaster clock, VLAN segmentation, IEEE 802.1Qbv time-aware traffic shaping

Arrow Lake Host (Intel i226 NIC)

TSN-capable inference host; clock synchronized to the switch via PTP

Camera(s)

Video source; supports either RTSP cameras (NTP/gPTP) or Basler GigE cameras (IEEE 1588v2 hardware PTP)

Traffic Injector

Runs iperf3 to generate background congestion and demonstrate TSN protection

This project demonstrates two complementary use cases for industrial edge AI, both using TSN infrastructure to protect latency-sensitive streams from background congestion:

Use Case 1 — Multi-Camera AI Inference with Deterministic Delivery#

RTSP camera streams from AXIS cameras are processed by DL Streamer for person detection. Inference results and simulated sensor telemetry are published over MQTT with PTP timestamps. An MQTT aggregation node measures end-to-end latency in real time, demonstrating how TSN protects critical streams from iperf3 background congestion.

Get Started — Use Case 1

Use Case 2 — SceneScape Multi-Camera Tracking with TSN and PTP#

Basler GigE cameras hardware-timestamp each frame with IEEE 1588v2 PTP. A patched GStreamer pipeline propagates these timestamps through DL Streamer into Intel® SceneScape for 3D multi-camera tracking. This use case measures how TSN congestion affects HOTA tracking accuracy and demonstrates that traffic shaping restores accuracy to baseline.

Get Started — Use Case 2

Documentation#

Key References#