Fall Detection using Wi-Fi Signals and Threshold-Based Activity Segmentation

We present the design and implementation of a low-cost, accurate and non-invasive wireless fall detection system utilising commercial off-the-shelf (COTS) 802.11n WLAN network interface cards (NICs). The system utilises the channel state information (CSI) of the wireless channel between a transmitter and a receiver. Notably, in addition to the CSI amplitude, the proposed system exploits the phase difference over 2 receiving antennas to detect patterns uniquely attributed to a human falling. Our extensive experimental results show that the CSI phase difference is a more granular measure at 5 GHz rather than the amplitude. The proposed method for fall detection consists of two stages. In the first stage, we quickly segment two types of actions, fall-like activities and falling activities to reduce the computational power required. In the second stage, we build a classification algorithm with newly defined features to detect three types of falls, namely walking-falls, standing-falls and sitting-falls. The concept of a sitting-fall is introduced whereby a person falls as they are standing up or sitting down. This is much more subtle than a walking-fall or standing-fall. To this end we introduce new features for signal classification such as the velocity of change of the standard deviation of the CSI phase difference. We also improve on existing features such as TimeLag proposed in [1]. We carry out extensive experiments to evaluate the performance of the proposed fall detection system. Particularly, the results demonstrate a balanced accuracy of 96% for the proposed system, compared to 91% for the top state-of-the-art solution [1].