Toward a Distribution Difference-Based Passive Motion Detection System Using WiFi Signals

Abstract

Passive motion detection (PMD) systems based on WiFi signals consider the traditional wireless network as a sensing network for detecting human motion. However, most WiFi-based PMD systems may perform poorly when the deployment environment changes (e.g., changes of transceiver or furniture placement). In addition, under the constraints of the limited devices and measurement noise, developing a robust and accurate motion detector is still heavily required. To this end, we propose R-MoDe to revisit the PMD problem from the perspective of the distribution difference of environmental statistics. First, R-MoDe exploits a statistical model based on temporal correlation function (TCF) to characterize the time-varying channel induced by human motion. Based on the statistical model, R-MoDe calculates environmental statistics by TCF and treats the distribution of the statistics under non-motion scenarios as a template profile, independent of the deployment environment and measurement noise. Next, with the limited devices, we regard available subcarriers as virtual sensors and design a clustering-vote-based sensor selection scheme to boost detection performance. Then, R-MoDe adopts the distance between the distribution of environmental statistics and the template profile as a robust detection metric. Finally, we propose a motion detection strategy based on double thresholds and hysteresis tracking to enable accurate and robust detection. R-MoDe is implemented on commodity WiFi devices in practical indoor environments. Experimental results show that R-MoDe can achieve an average motion detection rate of 96.19% and an average true negative rate of 97.04% using only one pair of transceivers, significantly outperforming the state-of-the-art methods.