WiFi CSI-based Fall Detection from Wearable Devices

As the population ages, falls increasingly pose a significant public health concern. This has led to the proposal of many fall detection methods. Some of these methods, based on Wi-Fi Channel State Information (CSI), offer a cost-effective alternative that protects user privacy, which may be compromised by some vision-based solutions. However, conventional CSI-based methods have a major limitation: both the Wi-Fi transmitter and receiver are statically deployed, and only users located within the range between them can be detected. To address this, we propose a solution where detection is based on users’ wearable devices, which send Wi-Fi data to the AP, and the AP collects CSI data for fall detection. We first conduct a case study to confirm its feasibility by considering both time and frequency-time domain features extracted from CSI data. We then apply two spectrogram-based CSI methods, image recognition and features classification on spectrograms. Practical evaluation shows their effectiveness with commercial off-the-shelf (COTS) devices, including Wi-Fi NIC and smartwatch. In particular, the image recognition method can achieve accuracy higher than 80% in most cases with mobile wearable devices.