Sequential Deep Learning for In-Home Activity Monitoring Using mm-Wave FMCW Radar

We present a stand-alone mm-wave radar system coupled with deep learning algorithms as the basis of an autonomous in-home free-living, physical activity recognition system. Using the mm-wave radar system, human spectrograms (time-varying micro-doppler patterns) are used to train deep Gated Recurrent Networks (GRU) to identify physical activities performed by a subject in his living environment. An overall model accuracy of 93% was achieved to classify five in-home physical activities. The system enables free-living human activity monitoring and gait cycle identification.