An ESP32-S3 sampling Wi-Fi channel state information 100 times per second, paired with a Raspberry Pi 4 for signal processing, can detect a motionless human by the chest displacement of breathing alone. This is wifisense-pi, an open-source project by The Masked Bear, and it works through walls using the 2.4 GHz signals already in the environment.
The system answers one question reliably: is something alive in this room. It cannot give position, and it cannot distinguish a 50 kg dog from a 50 kg human. Multiple occupants distort the signal more than one, but The Masked Bear confirms you cannot extract a reliable count. The one exploitable edge case: hold your breath and you disappear from the sensor entirely.
What makes the full project writeup worth reading is not the conclusion but the signal processing pipeline on the Pi side, and the specific CSI extraction approach on the ESP32-S3. Prior art exists, including earlier ESP32-based implementations, but this one is fully open-sourced on GitHub with a live dashboard. The architectural choices made to hit 100 Hz sampling without dropping packets are the real story here.
[READ ORIGINAL →]