RuView: Turning $9 WiFi Chips Into Contactless Vital Sign Monitors
Hook
A $9 ESP32 chip sitting in your hallway can detect your breathing rate through walls by measuring how your chest movements distort WiFi signal phase—and it never captures a single pixel of video.
Context
Privacy-invasive surveillance is the default architecture of modern smart homes. Ring doorbells upload your neighbors to Amazon's servers. Nest cameras require cloud subscriptions to function. Even 'local' AI security systems still process video frames, creating inherent privacy risks—what happens when that footage leaks, gets subpoenaed, or falls into the wrong hands?
RuView takes a radically different approach by exploiting a phenomenon called Channel State Information (CSI). When WiFi signals bounce around a room, they create interference patterns that encode spatial information about objects and movement. Your breathing chest compresses and expands the wavelength of 2.4 GHz signals by nanometers, shifting the phase relationship between transmitter and receiver. Traditional WiFi ignores this data—it's noise to be corrected. But extract those phase shifts, apply the right signal processing, and suddenly a $9 ESP32 becomes a contactless vital sign monitor. No cameras. No cloud uploads. No pixels to leak. RuView packages this academic research—previously requiring $5,000 software-defined radios—into firmware that runs on commodity IoT hardware you can buy at Adafruit.
Technical Insight
RuView's architecture is a three-layer pipeline that turns raw radio measurements into actionable sensor events. At the firmware layer, ESP32-S3 or ESP32-C6 chips capture CSI frames—amplitude and phase measurements across 52 OFDM subcarriers at 2.4 GHz or 64 subcarriers at 5 GHz. The firmware configures the Espressif HAL to expose these normally-hidden measurements and streams them via UDP at 100 Hz to avoid overwhelming the network stack.
The middle layer is where the magic happens. A Rust daemon (ruview-server) consumes these UDP packets and runs them through a signal processing pipeline. For breathing detection, it applies a bandpass filter isolating 0.1–0.5 Hz frequencies (6–30 breaths per minute), then computes the circular variance of unwrapped phase across subcarriers. The core insight is that chest movement creates coherent phase shifts across all 52 subcarriers simultaneously—random noise doesn't. Here's a simplified version of the breathing rate estimator:
use rustfft::{FftPlanner, num_complex::Complex};
fn estimate_breathing_rate(phase_samples: &[f32], sample_rate: f32) -> Option<f32> {
// Bandpass filter: 0.1-0.5 Hz (6-30 BPM)
let filtered = bandpass_filter(phase_samples, 0.1, 0.5, sample_rate);
// Zero-crossing count method for BPM
let mut crossings = 0;
for window in filtered.windows(2) {
if (window[0] >= 0.0) != (window[1] >= 0.0) {
crossings += 1;
}
}
// Convert zero-crossings to breaths per minute
let duration_seconds = phase_samples.len() as f32 / sample_rate;
let bpm = (crossings as f32 / 2.0) * (60.0 / duration_seconds);
// Sanity check: human breathing range
if bpm >= 6.0 && bpm <= 30.0 { Some(bpm) } else { None }
}
This zero-crossing approach is computationally cheap enough to run in real-time on a Raspberry Pi while maintaining sub-10ms latency. Heart rate detection uses the same pipeline but targets 0.8–2.0 Hz (48–120 BPM) and requires more aggressive noise reduction since cardiac signals are an order of magnitude weaker.
The real architectural innovation is the Cog plugin system. Instead of hardcoding every sensing algorithm into the firmware, RuView defines a WebAssembly interface where plugins receive CSI frames and emit JSON events. Each Cog module is a signed WASM binary under 100 KB that runs in a sandboxed Wasmtime runtime. The breathing detector is cog_breathing_v2.wasm, the fall detector is cog_fall_v1.wasm, and developers can write custom Cogs in any language that compiles to WASM. This architecture means you can deploy new sensing algorithms—occupancy prediction, gesture recognition, spatial mapping—without flashing firmware on dozens of ESP32 nodes scattered around a building.
The Home Assistant integration exposes each Cog's output as standard MQTT sensors. A single RuView node publishes to homeassistant/sensor/ruview_bedroom/breathing_rate with automatic discovery, making it a drop-in replacement for traditional presence sensors but with vital sign telemetry as a bonus. The ESP32-C6 variant adds WiFi 6 support and a low-power RISC-V coprocessor that gates CSI streaming—the radio only wakes the main CPU when motion is detected, dropping idle power consumption to 5 microamps.
The optional 'Cognitum Seed' edge device adds persistence and learning. It's a Raspberry Pi-sized board running a vector database that stores CSI fingerprints for each room. After a week of observation, it builds a kNN classifier that distinguishes 'kitchen occupied' from 'kitchen empty' patterns with 82% accuracy. Every CSI measurement gets an Ed25519 cryptographic signature and timestamp, creating an immutable audit trail—critical for elderly care applications where you need to prove a fall was detected at a specific time, not fabricated after the fact.
Gotcha
RuView's most serious limitation is that its headline feature—pose estimation—doesn't actually work. The repository includes cog_pose_v1.safetensors weights and marketing claims about DensePose-style body tracking, but the documented PCK@20 accuracy is 3.0%. That's effectively random guessing; clinical-grade pose estimation requires at least 35% PCK@20. The runtime code even hardcodes confidence=0 for pose predictions, suggesting the authors know it's not ready. This matters because the project's positioning around 'spatial intelligence' creates expectations that the underlying technology can't meet.
Vital sign accuracy is the second major unknown. While the signal processing theory is sound—breathing does modulate WiFi phase—the repository provides zero validation data. No comparison to medical-grade pulse oximeters. No multi-person experiments showing it can isolate one person's breathing from another's. No through-wall attenuation measurements. The documented range is '6–30 BPM', which conveniently excludes tachypnea (rapid breathing above 30 BPM) where actual medical diagnostic value lives. For elderly monitoring or sleep tracking, 20–30% error rates might be acceptable. For anything approaching medical use, this is a research prototype, not a product.
The 105-module Cog catalog is mostly vaporware padding. Modules like 'elevator-count', 'clean-room monitoring', and 'customer-flow analytics' appear in documentation but have no corresponding implementation in the repository. The advertised Cog registry at seed.cognitum.one/store returns 404. Only about 10 core modules—breathing, presence, fall detection—have actual code paths you can trace. This is common in open-source projects chasing GitHub stars, but it means you'll be writing your own Cogs for anything beyond basic occupancy detection.
Verdict
Use RuView if you're building privacy-conscious occupancy systems where cameras are legally or ethically unacceptable—elderly care facilities, medical offices, domestic violence shelters—and you can tolerate 20–30% error rates. The ESP32 firmware is production-quality with real power management and Home Assistant integration that actually works. The Cog plugin architecture is genuinely clever for composable sensing pipelines, and the honest benchmarking (rare in ML projects) tells you exactly what won't work. The ability to extract breathing rates from $9 commodity hardware without internet dependency is legitimately novel if you're experimenting with contactless health monitoring on a budget.
Skip RuView if you need clinical-grade vital signs (choose Xandar Kardian's FDA-cleared mmWave sensors), reliable pose estimation (the 3% accuracy is unusable), or multi-person disambiguation (completely untested). Also skip if you just need simple presence detection—plain PIR motion sensors or ESPresense Bluetooth tracking are simpler, cheaper, and more reliable. The technology is fascinating for researchers and hobbyists willing to write Rust to fill gaps, but it's not a drop-in replacement for commercial vital sign monitors despite the marketing implications.