A high-performance, real-time Human Sensing interface leveraging Wi-Fi Channel State Information (CSI). Utilizing an ESP32-S3 microcontroller as a dedicated ingestion node streaming 52-subcarrier complex per-packet IQ datasets, this Python and Electron hybrid desktop application natively infers activities across deterministic spatial timelines.
- Offline Data Generator: Construct synthetic data sequences locally to replicate deterministic motion behaviors for purely offline development environments.
- Dimensionality Processing: Automated outlier removals (Hampel Filtering), sanitization, extraction processing, dimensionality deduction (Principal Components Analysis), and frequency analysis (STFT) converting a massive 52-carrier tensor vector sequentially down to a readable 10-component tracking scope across continuous 2-second windows.
- Real-Time Dashboard Rendering: Electron-embedded interface updating Plotly traces for Spectrograms, Amplitude Heatmaps, and 3D PCA Tracking.
| Layer | Component Action | Functionality / Detail |
|---|---|---|
| Ingestion | Extractor (app.py, pyserial) |
Live deserialization from hardware Serial (ESP32) vs offline mock-data interpolation via Playback. |
| Processing | Sanitization & Aggregation (processing.py) |
Iterative operations: applies Hampel Filters, handles matrix conjugations, and STFT computations over 200 frequency segments. |
| Inference Inference | Long-Short Term Memory Analysis (ml_model.py) |
Employs PyTorch's backend parameter tensors csi_model.pth matched strictly into 4 classes mapping probabilities: [Unknown, Walking, Static, Falling]. |
| Front-End Visualizer | App Wrapper (index.js, Dash) |
Desktop instance containing interactive feedback variables such as live component tracking via Electron. |
- ESP32-S3 Firmware: Operable with
esp32-csi-tool. - Python: Version 3.8+ context dependencies (PyTorch, Dash, scikit-learn).
- NPM Frameworks: Electron.JS engine wrapper for Chromium.
Ensure you have your global NPM package manager and Python configured correctly contextually.
# 1. Install Node modules
npm install
# 2. Install Python core utilities
pip install -r requirements.txtGenerating mock variables will place resulting parameter values internally inside models/. For more details view its internal structure documentation.
# 1. Create a 20,000 instance Mock CSV
python src/generate_synthetic_csi.py
# 2. Compile Model Layer Logic & Save Data Dict parameters
python src/ml_model.py# Deploys Python backend daemon iteratively passing over local :8050 to UI
npm startDistributed under the MIT License. See LICENSE for more information.