You don't need cameras. You don't need wearables. You don't even need to see the room.
As long as there's a WiFi signal, you can know if someone is in the room, what they're doing, and even their heart rate and breathing frequency.
This isn't science fiction. It's an open-source project that has garnered 63,537 stars on GitHub: RuView (π RuView), developed by ruvnet, which added over 7,600 stars this week alone.
How Does WiFi Become a Sensor?
The principle is actually not that complicated.
Your WiFi router constantly broadcasts radio waves into the space. When someone walks into a room, sits down, moves around, or even just breathes, these waves are disturbed—the presence of a human body alters the signal's propagation path and phase.
What RuView does is capture these disturbances and convert them into usable data. Specifically, it leverages WiFi's CSI (Channel State Information)—physical layer data inherent in the 802.11 protocol that describes the amplitude and phase of the signal across each subcarrier.
Standard routers don't expose CSI, but the ESP32 chip does. RuView uses the ESP32 as a sensor node (costing as little as $9 each) to build a mesh network for collecting CSI data, which is then analyzed locally using a spiking neural network.
What Can It Do?
Presence and Occupancy Detection—Through-wall detection of whether people are present, how many, and entry/exit logs Vital Signs Monitoring—Contactless breathing rate and heart rate measurement, effective whether the subject is sleeping or sitting Activity Recognition—Walking, sitting, gestures, fall detection, extracted from temporal CSI patterns Environmental Mapping—RF fingerprinting to identify rooms, detect furniture movement, and discover new objects Sleep Quality—Continuous overnight monitoring, sleep stage classification, and sleep apnea screening
None of this requires cameras. No infrared sensors. No wearables. Just a $9 ESP32 board placed in the room.
The Hardcore Aspects of the Technical Architecture
RuView isn't just running a simple script. It features a fairly complete system design:
Multi-Frequency Mesh Scanning—Scans across 6 WiFi channels, and can even utilize neighbors' routers as "free radar illuminators." This approach is clever—you don't need to transmit your own signals; just leverage existing WiFi signals in the environment.
Spiking Neural Networks—Instead of traditional deep learning models, it uses Spiking Neural Networks (SNNs), which operate closer to biological neurons and run with extremely high efficiency on edge devices. The system can adapt to a new environment within 30 seconds.
Cryptographic Authentication—Every measurement is cryptographically authenticated via an Ed25519 witness chain, ensuring data integrity. This is essential for medical and security applications.
Edge Computing—The entire system runs completely on edge hardware, requiring no cloud or internet connection. Data never leaves the room.
63,000 Stars—This Number Is Insane
63,537 stars. What scale is that?
On GitHub, projects with over 60,000 stars are typically well-known frameworks (like Kubernetes or TensorFlow) or viral tools. For RuView, a relatively niche IoT/AI crossover project, reaching this number highlights a clear trend: the demand for camera-free sensing technology far exceeds expectations.
Consider the use cases:
- Elderly Care—No cameras (preserving privacy), yet capable of detecting falls and monitoring sleep
- Smart Buildings—Automatically detecting room occupancy to adjust HVAC and lighting
- Security—Through-wall intruder detection
- Healthcare—Contactless vital signs monitoring
- Sleep Research—Continuous overnight monitoring of breathing and heart rate
In these scenarios, "no cameras required" is a massive selling point in itself. Privacy-sensitive users and environments (bedrooms, nursing homes, hospitals) have a real and urgent need for camera-free solutions.
Honest Limitations
The project's README is remarkably transparent about its limitations, which is uncommon for GitHub projects:
ESP32-C3 and Original ESP32 Not Supported—Single-core performance is insufficient to handle CSI DSP processing. Requires an ESP32-S3 or newer model.
Limited Spatial Resolution with a Single Node—At least 2 nodes are recommended, or pairing it with the Cognitum Seed (their hardware product) yields the best results.
Limited Accuracy for Camera-Free Pose Estimation—Currently, PCK@20 is around 2.5% (evaluated using proxy labels). The team is working on camera-supervised training, targeting 35%+ PCK@20, but data collection and evaluation are still ongoing, with no publicly released metrics yet.
These aren't minor issues. Pose accuracy in particular—a 2.5% PCK@20 means that for most scenarios, precise human pose estimation using only WiFi signals isn't realistic yet. But for presence detection and coarse activity recognition, it's already sufficient.
How Does It Relate to "Smart Homes"?
Current smart home solutions typically detect people using either PIR sensors (only detect movement, can't distinguish humans from pets), cameras (privacy-invasive), or mmWave radar (high cost, high power consumption).
RuView offers a new alternative: leveraging existing WiFi infrastructure for sensing. You don't need to buy extra sensors—if you already have ESP32 devices, you can even deploy it at zero cost.
It might not be the next consumer-grade blockbuster, but in specific scenarios (elderly care, security, healthcare), its value is very real.
Primary Sources: GitHub - ruvnet/RuView, Cognitum Seed