AI-based Real-time Classification of Human Activity using Software Defined Radios

Real-time monitoring is an essential part in the development of healthcare monitoring systems. Research has shown that human movement affects the propagation of radio frequencies, as signals will reflect off the human body. Machine Learning techniques have been used in research to classify patterns observed in the signal propagation. This paper makes use of universal software radio peripheral devices to create a wireless communication link where the signal propagation data, known as channel state information, is collected while a user moves or remains still. A machine learning model which achieved an accuracy result of 93.25 % is used to classify between movement and no activity. Inference is then used to decide if the human position is sitting or standing and detected movements are used to differentiate between the two positions. The testbed implements cloud storage and a web-interface to present a visualisation of the human position.