RF / EM sensing/ Camera-free by construction

ZeroTouch

Sensing people with no camera in the room

Every Wi-Fi radio already measures how the room bends its signal. PRISM reads that from a $4 ESP32 to tell which zone you are standing in. AURA reads the electric field above your hand to move a cursor without a touch. Neither ever makes an image.

See it classify PRISM source AURA source

$4Radio, one ESP32

73.3%4-zone accuracy, unseen data

~10 msPer inference

0Cameras

PRISM · live zone display00:00.00

The live display, as someone walks between zones. The only sensor is the ESP32 on the cabinet.

§ 01

PRISM — Wi-Fi as radar

Passive RF-based indoor spatial mapping. A body moving through a room changes the multipath of every packet; PRISM learns what each zone’s change looks like.

Four filtered CSI heatmaps — empty room, zone A, zone B, zone C — each with a visibly different pattern across subcarriers over time. — Fig. 01 — Each zone leaves its own fingerprint across 52 subcarriers

Raw zone · confidence per window00:00.00

What the classifier prints, window by window, while the display above smooths it.

How the signal gets cleaned

Raw CSI amplitude is mostly noise. A Hampel filter removes outlier spikes, dynamic background subtraction deletes the static room, and a Butterworth band-pass keeps 0.1–3 Hz — the band where people move. What survives is a signal a 135-feature engine and a random forest can classify. Accuracy is 73.3% across four zones on held-out recordings, and 81.4% cross-validated on a two-zone corridor.

Four stacked plots of one subcarrier: raw noisy amplitude, after Hampel filtering, after background subtraction, and after a 0.1–3 Hz band-pass.

§ 02

AURA — a trackpad made of air

An MPR121 capacitive controller drives five electrodes. A hand above them shifts each channel’s field a little differently; a model maps those shifts to screen coordinates, and a Kalman filter keeps the cursor steady.

MPR121 · 5 chArduino · 2 Mbps serialbaseline − filtered RF / CNN-BiLSTMKalman / EMAcursor

5Electrodes

15×15Training grid

3Model backends

AURA on GitHub

Hover to bend the fieldSimulated

§ 03

Why no camera

Privacy from physics

Nothing to leak

A camera you promise not to misuse is still a camera. CSI and capacitance cannot be turned back into a picture of anyone.

Cost

Hardware already in the room

One $4 ESP32 listening to an ordinary 2.4 GHz router. No software-defined radio, no wearables.

Conditions

Dark rooms, other side of a wall

Radio does not need light, and passes through the drywall a lens cannot see past.