Real-Time Acoustic Radar
Device-free motion sensing with a laptop's speaker and microphone
Real-time motion detection from a laptop's own speaker and microphone: emit an inaudible 18–20 kHz tone and classify its Doppler-shifted reflections.
Overview
A research-oriented sensing system that detects human motion without cameras or dedicated sensors. A standard laptop speaker emits a continuous high-frequency tone (18–20 kHz) while the microphone records its reflections; motion shows up as Doppler shifts and multipath changes in the received signal.
The hard part is the data pipeline, not the model: full-duplex audio streaming with chunked, non-blocking capture, sliding-window buffering with overlap, band-pass filtering around the carrier, and STFT spectrogram features centered on the carrier frequency.
Detection is framed as binary classification over time windows — from a spectral-deviation threshold baseline to classical models (logistic regression, SVM, random forest) and a lightweight 1D CNN sized for real-time CPU inference.