ZeroTouch
Sensing people with no camera in the room.
Wi-Fi channel state as room-scale radar, and capacitive fields for hover at desk scale. No image exists, so none can leak.
What it does
Wi-Fi CSI radar on commodity ESP32 radios — no SDR required
Full CSI pipeline: Hampel filtering, background subtraction, 0.1–3 Hz band-pass
Random forest over 135 engineered features: 73.3% across four zones, ~10 ms per inference
Touchless cursor control from five capacitive channels, with a CNN-BiLSTM backend
Camera-free by construction, from room scale down to fingertip range
The long version
Camera-based sensing asks people to trust a promise. ZeroTouch removes the need for one: both halves of the system measure the electromagnetic environment directly, and neither produces anything that could be reconstructed into a picture of a room.
PRISM treats channel state information — the estimate every Wi-Fi radio already computes to equalise a signal — as a measurement of the space rather than a nuisance to correct. A body moving through a room perturbs multipath consistently enough to classify, once the pipeline removes outliers, subtracts the static room, and band-passes to the 0.1–3 Hz range where people move. A random forest over a 135-feature vector then names the zone — 73.3% across four zones on held-out recordings.
AURA covers the range where radio is too coarse. An MPR121 controller drives five electrodes, and a hand above them shifts each channel’s capacitance a little differently. The signal is small and drifts, so the pipeline works on baseline-minus-filtered error per channel; a random forest or a CNN-BiLSTM maps those errors to screen coordinates, and a Kalman filter steadies the cursor.
Together they span room-scale presence down to fingertip hover, and both run on commodity ESP32 radios rather than software-defined radio hardware.