Robust device-free fall detection using fine-grained Wi-Fi signatures

Abstract

Fall is one of the main threats to the health care of elderly living alone. If not timely treated, the elderly will be threatened with death. Traditional fall detection systems based on vision, sensor networks or wearable-device are either intrusive to user's daily life, or sensitive to the changing ambient environment. However, most of them have not fully taken the dynamic environment factors into account, which makes them un-robust and hinders them from being applied in practice. In this paper, we propose a robust and unobtrusive fall detection system using off-the-shelf Wi-Fi devices, which gather fluctuant wireless signals as indicators of human actions. Specifically, we design a lightweight classifier to eliminate the “bad antennas” in channel state information (CSI) so that we can extract features from the best CSI stream; by which, the negative effects aroused by the dynamic surroundings can also be removed. We also design a novel method to intercept the valid segment of signal of fall action by utilizing wavelet analysis and dynamic time window. Finally, we implement a full robust device-free fall detection system based on the proposed novel methods. In a typical indoor environment, the recognition accuracy for the fall is 91%, and the false alarm rate is only 0.06%. Experimental results show that our system is robust to the complex indoor radio frequency environments and achieves good performance.