Edge‑Based Anomaly Detection and Computer Vision for Critical Environments
Design and deploy a real‑time edge‑computing system that fuses multi‑sensor streams with computer‑vision models to detect chemical leaks, fires, and other hazardous anomalies, instantly alerting researchers and operators to prevent accidents.
Team Members:
Lucius Choi (Team Lead), Yu‑Hsien Wei, Cheng Chung, and Daniel Hsu
Project Deliverables:
- Edge‑AI Gateway Prototype: Advantech ARK‑1221L box configured with a Hailo‑8 accelerator and containerized inference pipeline.
- Noise‑Hardened Hardware Revision: Replace a plastic enclosure with a Faraday cage and migrate from breadboard to a custom PCB to minimize electrical noise.
- Multi‑Sensor Dataset: Time‑synchronized recordings of gas, temperature, humidity, spectral, and CO₂ readings with labeled events.
- Computer‑Vision Module: YOLOv8‑based fire/leak detector optimized for Hailo‑8 acceleration.
- Anomaly‑Detection Model: Isolation Forest + Autoencoder hybrid with automatic thresholding and dashboard‑ready alerts.
- Web Dashboard: Live status, trend graphs, and push‑notification integration.
- Technical Report & Slide Deck: Comprehensive documentation of architecture, benchmarks, and comparison against a baseline cloud‑only solution.
