PROJECT / 002
Fog-Based Multi-Node Environmental Anomaly Detection on ESP32
Sub-200ms fire & intrusion detection, entirely on ₹2,356 of hardware.
Overview
Co-authored research paper proposing a three-tier fog-computing safety system built entirely on commodity ESP32 microcontrollers, delivering cloud-free, sub-200ms fire and intrusion detection.
My Role
Co-author — contributed to system architecture, the dual-modality decision engine, and experimental evaluation.
Status
Research
The Problem
Cloud-mediated IoT safety systems can take seconds to round-trip a decision — too slow for fire or intrusion response — and single-sensor thresholding produces high false-positive rates from cooking fumes, aerosols, or passing animals.
The Solution
- Designed a context-aware dual-modality sensor-fusion algorithm — gas + luminosity for fire, PIR + acoustic for intrusion — requiring two independent physical channels to agree before triggering an alarm.
- Calibrated operating thresholds via an F1-maximizing sweep over 1,500 labeled observations rather than datasheet defaults.
- Built a three-tier architecture: two edge sensor nodes streaming telemetry over local Wi-Fi to a static-IP ESP32 fog server that aggregates, decides, actuates, and hosts a self-refreshing dashboard — no cloud dependency at all.
Results & Metrics
95.0% / 95.2%
F1-score (fire / intrusion)
0.974 / 0.981
AUC-ROC
65% (17% → 6%)
False-positive reduction
180ms avg (13.3x faster than cloud MQTT)
End-to-end latency
97.3% over 60 min
Wireless packet delivery
₹2,356 (~$28)
Total hardware cost
Next project
Smart-City Traffic Monitoring