RestAware: Non-Invasive Sleep Monitoring Using FMCW Radar and AI-Generated Summaries

RestAware is a comprehensive framework for smart and non-invasive sleep monitoring using a 24GHz FMCW radar sensor. This repository contains raw data collection, preprocessing, posture classification, exploratory analysis, and sleep summary generation using large language models (LLMs). It supports healthcare researchers and developers in building intelligent and privacy-preserving sleep monitoring system.

---

Repository Structure

Dataset/

• Description: Contains all raw and decoded sleep data.
• Contents:
  • radar_*.csv: Unprocessed sleep monitoring data captured via radar sensor for each participants.

EDA/

• Exploratory Data Analysis of decoded data.
• Generates:
  • Sleep posture and movement visualizations
  • Per-person statistical summaries (data_summary.txt)
  • Visual assets (EPS format)

Posture_classification/

• Goal: Classify sleep postures using machine learning and neural networks.
• Contents:
  • ML_classifier/classification.py: Compares multiple ML models using LazyPredict.
  • NN_classifier/classification.py: Uses a PyTorch neural network to classify sleep posture.

sleep_summary_generation/

• Goal: Automatically generate natural language sleep summaries from posture and movement data.
• Models Used:
  • mistral.py: Mistral 7B Instruct
  • llama.py: LLaMA-2 7B Chat
  • falcon.py: Falcon-7B-Instruct
• Outputs high-level sleep summaries.

physical_parameters/

• Purpose: Scripts to interface with and collect data using a 24GHz FMCW radar sensor.
• Includes:
  • Radar initialization and configuration
  • Data capture logic
  • File generation for raw sleep data

---