A Contactless Fall Detection Method Based on CSI Signals

Abstract

Wi-Fi channel state information (CSI)-based human behavior sensing technology has been widely used in areas like elderly care and smart homes. Given that falls are particularly dangerous for the elderly, traditional detection methods are costly and pose privacy risks, and existing CSI-based fall detection algorithms don' t work well in small, high-risk areas like bathrooms, this paper proposes a lightweight Wi-Fi fall detection method called Wi-FD. This method selects effective antennas through sliding window variance, uses combined filtering to reduce noise, applies principal component analysis to select optimal subcarriers and generate feature maps; behavior features are extracted via fuzzy C-means segmentation and local binary patterns, and a behavior fingerprint library is built using support vector machines to achieve real-time fall detection. Experiments in multiple scenarios show that the proposed method has an identification accuracy of 92.3%, is highly real-time, doesn ' t need visual capture, and poses no privacy risks. It can adapt to complex, small home environments and is suitable for smart fall monitoring for the elderly, making it quite practical.