Can Wi‑Fi Turn Your ESP32 Into a Camera‑Free Human Pose Detector?

Project Overview

RuView (formerly WiFi DensePose) is an open‑source edge‑AI perception system that implements Wi‑Fi‑only human pose estimation, vital‑sign monitoring, and presence detection based on the Carnegie Mellon University paper “DensePose From WiFi” (arXiv:2301.00250). The implementation runs on a single low‑cost ESP32‑S3 chip (≈ US$8) and has been rewritten in Rust for high performance.

Technical Principles

The system exploits Wi‑Fi Channel State Information (CSI), which captures amplitude and phase changes of each sub‑carrier caused by human bodies (reflection, diffraction, and phase disturbance). The processing pipeline is:

Collect raw CSI from the Wi‑Fi receiver.

Extract amplitude and phase variations per sub‑carrier.

Feed the processed CSI into a deep‑learning model that maps the signal perturbations to 3‑D human pose, motion, and physiological signals.

Key technical details:

The Rust implementation provides roughly an 800× speedup over the original Python version, enabling real‑time inference on edge devices.

Inference throughput is reported at 54 000 frames per second for pose estimation.

Vital‑sign extraction uses a band‑pass filter followed by Fast Fourier Transform (FFT) peak detection to derive breathing rates (6–30 BPM) and heart rates (40–120 BPM) from CSI phase dynamics.

Presence detection relies solely on RSSI values, achieving sub‑millisecond latency (< 1 ms) on standard home routers.

Core Capabilities

Wall‑Penetrating Pose Estimation : Detects and renders skeletal poses through solid walls up to 5 m depth, supporting multi‑person tracking.

Contact‑Free Vital‑Sign Monitoring : Measures breathing and heart rate without wearables, using CSI‑based signal processing.

Ultra‑Fast Presence Detection : Determines occupancy from RSSI with latency < 1 ms.

Deployment Guide

A minimal, hardware‑free demo can be run via Docker:

docker pull ruvnet/wifi-densepose:latest

docker run -p 3000:3000 ruvnet/wifi-densepose:latest

After the container starts, open http://localhost:3000 in a browser to view the visualization panel.

For full functionality the recommended hardware consists of 3–6 ESP32‑S3 modules forming a mesh network (total cost ≈ US$54). Research‑grade NICs such as Intel 5300 or Atheros AR9580 can also capture CSI, whereas ordinary Wi‑Fi adapters provide only RSSI and cannot support pose or vital‑sign inference.

Deployment challenges include a per‑room calibration step (“calibration hell”) that must be performed to align CSI signatures with the physical environment.

Community Feedback and Risks

Independent audits on public forums have raised concerns that:

Core functionalities may be incomplete or untrained, with some reported data appearing synthetic.

No third‑party verification has reproduced the full pipeline using the specified hardware.

Technical compliance risk: Passive Wi‑Fi sensing that penetrates walls may conflict with personal data protection regulations; the IEEE 802.11bf perception standard is still under development.

Usage risk: The project is in an Alpha stage; APIs, firmware, and documentation change frequently. No pre‑trained models are provided, requiring users to train models from scratch. Known issues include inaccurate multi‑person counting and convergence difficulties.

Conclusion

RuView demonstrates a research‑grade implementation of Wi‑Fi‑based perception with impressive performance metrics and a convenient Docker deployment. However, the lack of independent validation and reported implementation gaps make it more suitable for academic exploration and prototype development rather than production or safety‑critical applications.

https://github.com/ruvnet/RuView