layout	width title description tableOfContents outline pagination metadata wide visible true visible visible true visible true visible true visible

GUARDEN - Garden Urban AI Rodent Detection and Environment Node

image [1] logo

Seattle University - College of Science and Engineering

ECEGR 4640: Internet of Things (IoT)

Ezekiel A. Mitchell
[email protected]

Danny Woo Community Garden (Andy Allen, [email protected])

---

Overview

The Danny Woo Community Garden (established 1975) is a 1.5-acre urban P-Patch serving Seattle's International District/Chinatown community. Rat infestations significantly reduce harvest yields. GUARDEN offers a data-driven solution utilizing distributed AI-powered camera nodes to monitor and map rat activity, facilitating targeted pest control.

---

Key Features

• Edge AI Detection - On-device TensorFlow Lite inference on ESP32-C3
• Low Power Operation - Motion-triggered wake with deep sleep (10µA standby)
• Privacy-First - Metadata-only transmission, no images leave devices
• Real-Time Monitoring - Local web dashboard with live visualization
• Data-Driven Insights - Activity heatmaps, peak detection times, nesting area identification
• Offline Operation - Fully self-contained system with no internet dependency
• Local Network - Pi 5B+ acts as a WiFi access point for ESP32 nodes and user access

---

Technical Specifications

Component	Specification	Performance	Power
Processor	ESP32-C3 RISC-V @ 160MHz	<1s inference target	240mA active
Camera	OV2640 2MP sensor	96×96 grayscale capture	80mA
AI Model	MobileNetV2 α=0.35 TFLite INT8	>65% accuracy target	160mA
Motion Sensor	PIR sensor	Event-driven wake	50µA standby
Sleep Mode	Deep sleep	Extended battery	10µA
Connectivity	WiFi 802.11 b/g/n + MQTT	Local AP only	120mA TX
Storage	MicroSD + PostgreSQL	Offline resilience	20mA write
Power	Solar + 5000mAh LiPo	10+ day runtime	Self-sustaining

---

Performance Targets

Metric	Target	Status
Detection Latency	<1s	In progress
Battery Life	3+ days	In progress
Detection Accuracy	>65%	In progress
Network Uptime	>20%	In progress
Storage Capacity	3+ days	In progress

---

Project Structure

GUARDEN/
├── edge/                        # ESP32-C3 firmware (PlatformIO)
│   ├── platformio.ini
│   └── src/
│       ├── main.cpp             # Entry point, deep sleep logic
│       ├── camera.cpp/.h        # OV2640 init + capture
│       ├── detector.cpp/.h      # TFLite inference wrapper
│       ├── mqtt_client.cpp/.h   # WiFi + MQTT publish
│       ├── model_data.h         # Auto-generated from quantize.py
│       └── config.h             # Pin defs, thresholds, node ID
│
├── fog/                         # Raspberry Pi hub (all services)
│   ├── broker/mosquitto.conf    # MQTT broker config
│   ├── ingestion/main.py        # MQTT → PostgreSQL ingestion
│   ├── dashboard/app.py         # Flask dashboard + API
│   ├── db/schema.sql            # Database schema
│   ├── setup.sh                 # One-shot Pi setup script
│   └── requirements.txt
│
├── model/                       # ML pipeline (run on laptop)
│   ├── train.py                 # MobileNetV2 training
│   ├── quantize.py              # INT8 TFLite export
│   ├── evaluate.py              # Accuracy benchmarks
│   └── guarden_v1_int8.tflite   # Trained model
│
├── config/
│   ├── nodes.yaml               # Node locations + GPS coords
│   └── .env.example             # Environment variable template
│
└── tests/
    ├── test_ingestion.py
    ├── test_dashboard.py
    └── test_model.py