Sensing without a camera in the room

RuView turns ordinary WiFi into a spatial sensor. Every router already floods a space with radio waves, and when people move, breathe, or sit still, they disturb those waves measurably. RuView captures those disturbances as Channel State Information from low-cost ESP32 sensors and turns them into presence detection, breathing and heart-rate estimation, activity recognition, and pose, through walls, in the dark, with no cameras or wearables. For anyone who wants occupancy and wellness sensing without putting a camera or a phone app in a bedroom or a care facility, that privacy posture is the entire appeal.

It runs on the edge by design. A node can be a $9 ESP32 board, the published model is 4-bit quantized to fit in 8 KB and run in microseconds on a Raspberry Pi, and the system learns each environment locally with no cloud, no internet, and no video.

What it can sense

The README is specific about capabilities and their methods, which is unusual for this space:

  • Presence and occupancy, including counting and entry/exit tracking, with a phase-variance fallback that needs no model at all.
  • Vital signs: breathing rate via a 0.1 to 0.5 Hz bandpass on wrapped phase, and heart rate via a 0.8 to 2.0 Hz band, both contactless and real-time.
  • Activity and fall detection from temporal patterns, with fall detection gated by a phase-acceleration threshold plus debounce and cooldown to limit false alarms.
  • Pose estimation, 17 keypoints, from a model published on Hugging Face.

It exposes 21 entities per node, eleven raw signals plus ten inferred semantic states such as someone-sleeping, possible-distress, bed-exit, and elderly-inactivity-anomaly, which is what makes it useful to a home rather than just a research demo.

How it fits a smart home

This is where RuView is more product than paper. It drops into Home Assistant with a single --mqtt flag via an MQTT publisher, and pairs into Apple Home, Google Home, and Alexa as a discoverable Matter bridge, so Siri, Google Assistant, and Alexa can report presence and vitals by room with no custom skills. It ships starter Home Assistant blueprints to get automations going. The hardware side is an ESP32 mesh that scans across six WiFi channels and can use neighboring routers as free radar illuminators, with every measurement cryptographically attested via an Ed25519 witness chain.

The honesty that should raise your confidence, not lower it

Most projects in the sensing-hype space quote a suspiciously round accuracy number. RuView does the opposite, and it is worth calling out. The README states that an earlier “100% presence” figure was measured on a single-class recording and has been retracted, replaced with an honest, label-free held-out temporal-triplet accuracy of 82.3 percent for the v2 encoder, up from 66.4 percent raw. Its pose model reports 82.69 percent torso-PCK on the matched MM-Fi protocol, with the benchmark method named and auditable. A project that retracts its own inflated claim and publishes the harder, lower honest number is signaling the kind of rigor you want before trusting a sensor with fall detection. Treat the capabilities as real but environment-dependent: RF sensing needs per-room calibration (around 30 seconds) and accuracy varies with layout.

Where it fits, and the maturity note

Reach for RuView if you want contactless presence and wellness monitoring that is private by construction and cheap to deploy, especially for elder care, sleep monitoring, or occupancy automation where a camera is unacceptable. The smart-home integrations make it usable today rather than a lab toy.

Set expectations on maturity. It uses build-numbered releases (v1613 in June 2026) and has 198 open issues as of 2026-06, and the physics means results depend on your environment. Open-source WiFi sensing on commodity hardware has mostly lived in academic work like DensePose-from-WiFi research; an MIT-licensed, ESP32-based, smart-home-integrated implementation at this scale is genuinely unusual, so there is no mature equivalent to compare it against on equal terms. Pilot it in your actual space before relying on it for anything safety-critical.

RuView is written in Rust; for more of what is moving in that ecosystem, see Rust trends. For what else is climbing across GitHub, see the daily digest and the weekly report.

FAQ

Does RuView use cameras? No. It reads WiFi Channel State Information from ESP32 sensors, so it works through walls and in the dark without cameras or wearables.

What hardware do I need? ESP32 nodes, as low as $9 each, optionally paired with a small host like a Raspberry Pi for inference. The quantized model fits in 8 KB.

How accurate is it really? The project honestly reports 82.3 percent held-out accuracy for its v2 encoder and retracted an earlier 100 percent claim. Accuracy is environment-dependent and needs brief per-room calibration.

Does it work with Home Assistant? Yes, via a single --mqtt flag, and it pairs with Apple Home, Google Home, and Alexa as a Matter bridge.