A contactless sensing system for indoor fall recognition based on channel state information

This paper introduces the designs and the implementation of a non-invasive indoor fall recognition system based on channel state information (CSI) in the Wi-Fi physical layer. We use a wireless router and a laptop computer equipped with an Intel Wi-Fi Link 5300 network card (802.11n) to setup a hardware platform. The platform receives and stores CSI data under various circumstances when a person in the Wi-Fi covered area stands up, sits down, walks, and falls. The CSI data are then processed and analysed using Matlab tools. Feature variables such as signal offset strength, period of motion, normalised standard deviation, median absolute deviation, interquartile range, and signal entropy are examined and best feature variables are chosen. Finally, cross validation algorithm and support vector machine (SVM) are used to establish the pattern recognition model. We tested the system in a laboratory environment and the experimental results showed that the fall incidents were effectively differentiated from other movements.