‘Seeing’ through walls with WiFi: what can you actually detect?
WiFi can detect presence and estimate movement without a camera. How does this technique work, and what are its limitations?
The body leaves a trace in the signal
Detecting someone behind a partition without filming the room naturally attracts attention. WiFi sensing uses changes in radio waves to infer what is happening in a space. ‘Seeing’ here means an estimate. The system does not produce a photograph of the person.
Espressif, the ESP32 manufacturer, describes CSI (Channel State Information) as a measurement of radio channel characteristics, including signal amplitude and phase. When someone moves, they change propagation. In a sufficiently quiet environment, subtler movements such as breathing may also leave measurable variations.
The chain therefore starts with hardware capable of retrieving these measurements. Signal processing then looks for useful variations. Depending on the task, a trained model may learn to connect this data to presence or posture. Having a home WiFi network is insufficient to obtain all these functions.
A research direction predating RuView
The paper DensePose From WiFi, submitted on 31 December 2022 by Jiaqi Geng, Dong Huang and Fernando De la Torre, explores human pose estimation from WiFi. The researchers use amplitude and phase to train a neural network. Their protocol pairs WiFi measurements with images serving as references during training; prediction then uses the radio signal.
The authors also describe failures: underrepresented poses and multiple people present simultaneously complicate reconstruction. Performance drops when the tested layout differs from training environments. A demonstration in one room therefore does not guarantee the same result in another home.
Breathing and heart rate are the subject of other work. A publication by Yu Gu and co-authors, submitted on 14 August 2019, presents a WiFi prototype for monitoring during sleep. The authors pay particular attention to antenna placement to detect these small movements. These results concern their experimental setup.
What the RuView repository offers
The open-source RuView project brings together CSI capture tools and processing for presence detection, breathing signals and other uses. Its documentation, consulted on 19 October 2026, distinguishes several hardware setups and pipelines. Some configurations can run locally without Internet access.
On the ESP32, the firmware documents a presence indicator based on signal variations, along with experimental breathing and heart-rate estimates. It states that this embedded processing runs no trained neural model and performs no pose estimation. Pose estimation belongs to separate processing on a host machine.
The README separates dataset results from live pose reconstruction, which remains limited depending on the model. The Docker demo can run on simulated data. Another demonstration combines a webcam and CSI: the displayed animation alone therefore does not prove that every joint is reconstructed using only WiFi.
The firmware also warns of possible false positives from radio interference. Its heart-rate estimates are experimental and are not medical measurements. No RuView hardware trial was carried out for this article.
A partition does not become transparent
RuView describes through-wall detection as dependent on the signal. Read this capability alongside its conditions: a usable radio link and correctly positioned sensors. The phenomenon’s existence provides neither universal range nor identical accuracy behind every obstacle.
‘Without a camera’ may refer to the usage phase even if images helped train a model. ‘Without Internet’ still requires a radio source and receivers. To understand a demonstration, I would first examine where the displayed data comes from and which machine runs each process.
From household presence to sleep monitoring
One possible application would control lighting according to room occupancy. Sleep monitoring or fall alerts also attract interest. These uses require different validation: a light switched on mistakenly and a missed fall do not have the same consequences.
To evaluate a prototype, I would start with a limited question, such as distinguishing an empty room from an occupied one. Results should be compared with a reference, errors counted and tests repeated after moving furniture. Patient monitoring would require evidence suited to that use; a number displayed in an interface is insufficient.
No image does not resolve privacy
Local data processing reduces transfers to a remote service. Yet a presence history can reveal schedules and habits. Physiological estimates may be sensitive even without recorded video. The question also concerns inferred information and who can access it.
I would want to know when this type of sensor is running, which area it observes and how to disable it. Before installation, agree with the people concerned which data will be retained and for how long. Making software public helps examine how it works; it does not guarantee how it will be used.
To explore further, the repository and publications below let you compare promises with protocols. In a first trial, identifying a real CSI source and documenting observed errors would already be more informative than an animated silhouette on screen.
Sources and useful links
- Espressif, ESP-CSI: CSI principles and capture (consulted on 19 October 2026)
- Geng, Huang and De la Torre, DensePose From WiFi (31 December 2022)
- Gu et al., WiFi-based Real-time Breathing and Heart Rate Monitoring during Sleep (14 August 2019)
- RuView: hardware, demonstrations and model maturity (consulted on 19 October 2026)
- RuView, ESP32 firmware: embedded processing and limitations (consulted on 19 October 2026)