Abstract
Conference Title: 2018 IEEE 87th Vehicular Technology Conference (VTC Spring) Conference Start Date: 2018, June 3 Conference End Date: 2018, June 6 Conference Location: Porto, Portugal This paper proposes a method of recognizing and classifying the basic activities such as forward and backward motions by applying a deep learning framework on passive radio frequency (RF) signals. The echoes from the moving body possess unique pattern which can be used to recognize and classify the activity. A passive RF sensing test- bed is set up with two channels where the first one is the reference channel providing the un- altered echoes of the transmitter signals and the other one is the surveillance channel providing the echoes of the transmitter signals reflecting from the moving body in the area of interest. The echoes of the transmitter signals are eliminated from the surveillance signals by performing adaptive filtering. The resultant time series signal is classified into different motions as predicted by proposed novel method of convolutional neural network (CNN). Extensive amount of training data has been collected to train the model, which serves as a reference benchmark for the later studies in this field.