Human Activities Classification in a Complex Space Using Raw Radar Data

The classification of human activities through the utilisation of radar mainly focuses on the analysis of the time-frequency domain, typically through spectrograms obtained with Short-Time Fourier Transform. Commonly the original information format in the raw complex radar data is being ignored. In this work, we propose and evaluate a new recurrent neural network architecture to decode the time sequence of raw radar data over a longer time than previously attempted, through training in a complex space of in-phase and in-quadrature data. We explore a solution for sequence analysis problems when adapting a recurrent neural network; our best network architecture exhibits a significant accuracy in performance for over seven participants with six activities, collected over 60 seconds.