Human Activity Recognition System Using WiFi Sensing and Deep Learning

Human activity recognition systems can be used for various purposes such as monitoring, authentication, and telemedicine. In this research, a non-invasive, high privacy, easy to implement, and affordable human activity recognition system based on WiFi and deep learning is developed. Sixteen activities; including upper body, lower body, and whole body movement; were recognized by utilizing Channel State Information (CSI) contained in the WiFi signal. Measurements were carried out in an empty room with dimensions of 6*8 m with the distance between the transmitter and receiver being 1, 3 and 6 meters from the subject. Google Teachable Machine is used to recognize activities carried out. From the measurement result, the accuracy shows more than 97%. It is also evident that the further the measurement distance, the worse the recognition results. This is due to the increasing amount of noise in the radio channel as the distance increases.