Project

Adaptive Sensor Fabric (ASF)

Edge-first, sensor-agnostic AI middleware for urban intelligence.

Context

This project is a proposal submission for the Department of National Defence IDEaS Program — Challenge CFP006 - CH15: "Turning Urban Data into Real-Time Insight through AI". The system demonstrates a working prototype built for edge deployment and offline operation.

What It Is

ASF is a middleware platform that converts heterogeneous urban data sources (traffic cameras, weather stations, AIS marine traffic, ADS-B aircraft, air quality monitors, RF sensors, transit systems) into explainable, confidence-scored situational awareness alerts. It does not require new hardware — it sits on top of existing city infrastructure.

Key Capabilities

5+ Sensor Modalities

Visual/optical, radio frequency, environmental, position/tracking (ADS-B, AIS), maritime — via 44+ adapters.

7-Engine Fusion Pipeline

Sensor Intelligence → Knowledge → Context → Traffic → Fusion → Confidence → Decision Intelligence.

Explainable AI

Every alert includes evidence chain, contributing sensors, confidence percentage, and plain-language operator recommendation.

Edge Deployable

Single Docker container, no cloud dependency, runs on laptop or NVIDIA Jetson. Terminal dashboard works without a browser.

Demo: Try It Yourself

The system runs entirely offline with simulated data. No API keys, no internet connection, no databases required.

Quick Start

cd asf-poc
pip install -r requirements.txt
python3 terminal_dash.py        # Terminal dashboard, no internet
python3 dashboard.py --trial     # Web map + terminal UI

See the full Demo Procedure for a complete walkthrough with 5 demo acts (15 minutes total).

Live Map

Below is the interactive ASF map showing Halifax-area zones, observations, and fusion events. The map server runs alongside this site.

Open full-screen →

Architecture

All sensor inputs normalize to a common Observation schema (timestamp, location, sensor_type, confidence, classification). The 7-engine pipeline processes observations through successive layers of abstraction, producing Assessment objects with provenance tracking back to source observations.

Pipeline Flow

Sensor Inputs → Observation Schema → SIE → Knowledge → CIE →
Traffic → FIE → Confidence → DIE → Alerts

Technical Details

  • Codebase: ~8,000 lines of Python across 117 source files
  • Tests: 26 automated tests, all passing
  • Protocols: MQTT, REST/HTTP, WebSocket, RTSP/RTMP, file-based
  • Geofencing: 9 Halifax zones with configurable centers and radii
  • Fusion: Temporal (500ms window) + spatial (200m radius) — JDL Level 1-2
  • Deployment: Docker (slim Python 3.12, 450MB image)
  • Live integrations: ECCC weather, AQHI, USGS earthquakes, OpenSky ADS-B, AISHub AIS, Blitzortung lightning, CWFIS wildfires, CHS tides, Halifax GTFS-RT transit

Privacy & Compliance

Privacy-by-design architecture: no PII collection, public data only, edge processing keeps raw data local, structured classifications only. Designed to align with PIPEDA and applicable provincial privacy legislation. Designed for the DND IDEaS program's Essential Outcome #4.

Interested in the technical details?

The full proposal, architectural documentation, and demo procedure are available.

View on GitHub Contact Me