An intelligent, distributed digital ecosystem designed to integrate IoT technologies into natural environments. The Smart Park project focuses on the Sila Grande case study, providing hikers, athletes, and tourists with real-time health monitoring, activity recognition, and safety alerts while transforming the park into a connected, intelligent ecosystem.
The system follows a Cloud-Edge paradigm where devices and software modules collaborate to provide seamless services even in remote areas where network connectivity is intermittent or absent.
- Edge Processing (Smartphone): Acts as the central coordinator, running Machine Learning models locally for immediate feedback, reducing network traffic by processing raw data on-device, and providing offline persistence via local storage.
-
Sensor Suite:
- Shimmer Sensor: Produces a continuous accelerometric stream at 50 Hz for activity recognition.
-
Smart Ring: Performs asynchronous measurement of physiological parameters, including Heart Rate,
$\text{SpO}_2$ , and Blood Pressure.
- Cloud Infrastructure (AWS): A serverless backend using AWS Lambda for data processing and Amazon DynamoDB for persistent, scalable storage, protected by HTTPS/TLS and API keys.
- Prototype Interface: An ESP32-based Web Server provides remote access to historical data for logged users, ensuring multi-workstation accessibility.
- ML Pipeline: Uses a CNN model running on TensorFlow Lite, where data is segmented into 100-sample windows (2 seconds) with a 50% overlap for high-accuracy inference.
- Classification: Distinguishes between Moving and Not-moving, with specialized recognition for Walking, Sitting, Standing, and Jogging.
- Multi-Parametric Validation: Correlates physiological data (e.g., heart rate) with activity context (e.g., running vs. resting) to minimize false positives (e.g., detecting tachycardia at rest).
- Fail-Safe Mechanism: Features an "Emergency Overlay" that wakes the display and provides a visual countdown window, allowing users to manually dismiss false alarms before emergency synchronization is triggered.
- Shimmer Optimization: Raw signals are cached in volatile RAM and summarized into X, Y, Z means. An atomic flush every 5 minutes sends JSON batches to AWS, significantly reducing network traffic, energy consumption, and cloud costs.
| Component | Technology |
|---|---|
| Mobile OS | Android 8.0 (API 26+) |
| Edge Intelligence | TensorFlow Lite (CNN-LSTM architecture) |
| Communication | Bluetooth Low Energy (BLE) |
| Local Storage | SQLite (isolated per UID) |
| Cloud Backend | AWS Lambda, DynamoDB, API Gateway |
| IoT Web Interface | ESP32 (WiFi, ArduinoJson, WebServer) |
- Multi-User Isolation: Firebase Authentication generates a unique UID for each user, creating an isolated local SQLite sandbox.
- Privacy Adherence: Personal data remains on the device; cloud transmissions are anonymized and associated only with a UID.
- Resource-Constrained Optimization: Designed to minimize CPU and battery usage through batching, careful thread management, and the avoidance of continuous raw data streaming.
- Expanded ML: Integration of activity-specific models (climbing, cycling, steep slope descent).
- Adaptive Profiling: User-configurable "Energy-vs-Latency" profiles to balance battery life during extended excursions.
- iOS Porting: Expanding ecosystem accessibility to Apple devices.
- Advanced Anomaly Detection: Specialized edge models for cardiac arrhythmia and fall detection.
Developed by: Desando Francesco, Marzano Mario, Masdea Miriam, Turano Edoardo.