Wi-Fi Sense: A Privacy-Preserving Wi-Fi CSI Motion Detection Prototype for Smart Building Automation and Security

Abstract

Sensing for indoor motion currently relies on passive infrared (PIR) sensors or camera-based systems, each of which present critical limitations such as missing stationary occupants or serious privacy concerns. This paper present Wi-Fi Sense, a device-free motion detection prototype built on two ESP32-S3 boards that communicate over ESP-NOW on a fixed channel. It exploits CSI amplitude shifts caused by human presence, captured across approximately 108 orthogonal frequency-division multiplexing (OFDM) subcarriers per frame and processed offline through Hampel outlier filtering, baseline normalisation, and Non-consecutive Best Variance Index (NBVI) subcarrier selection. A Moving Variance Segmentation (MVS) classifier then compares the rolling variance score against a threshold drawn from a still-room baseline recording, requiring no labelled data for operation. Testing consisted of five occupancy conditions in which MOTION classification rates came out to 55.4%, 52.8%, 75.2%, and 97.7% for the one-, two-, three-, and four-person trials, with no detections in the unoccupied baseline from which the threshold was derived. Detection is comparable at one and two occupants and improves markedly at three and four, indicating that a threshold drawn from a still-room recording is usable without per-environment training at moderate occupancy. Beyond the laboratory trials, these properties point to direct use in energy-aware HVAC and lighting control, occupancy-triggered security alerts, and continuous monitoring in healthcare or residential settings, where the two-board footprint and short calibration window make building-wide deployment realistic without the camera infrastructure or privacy trade-offs such environments would otherwise require.