The population of elderly people living independently is at risk of falling while carrying out activities of daily living. These falls can have physical or psychological consequences that affect the quality of life of these people if they are not treated promptly. Wireless perception and detection of these events through radio frequency (RF) signals is an approach that has gained popularity because it does not require the carrying of sensors or video monitoring. In this work, we present the main considerations for the development of a radio sensing platform using RF signals with characteristics similar to WiFi. In addition, we performed a proof of concept with actual falls to measure and compare frequency dispersions, also known as Doppler signatures, caused when a person interferes with RF signals. The platform was able to achieve a maximum accuracy of 92% and a false-negative rate of 8% using a support vector machine algorithm that is a classifier in a feature-based approach. The results show the importance of architecture with which such platforms are designed and the key techniques for their operation.
Key Words: Fall detection, Doppler signatures, WiFi, Polarization, Spectrogram.
;)
