Towards Real-Time Implementation for the Pre-Processing of Radar-Based Human Activity Recognition

The rapid aging of the population combined with the correlation between age and the increase in falls pushes us to create new ways to monitor the elderly. The privacy of radar data can respond to one of the weaknesses of existing technologies, but the huge amount of radar data to process becomes a challenge to process. We therefore introduce a first architecture allowing the processing of its data in real time. The radar technology used is an off-the-shelf Frequency Modulated Continuous Wave radar Ancortek (SDR 980AD2). It is followed by a pre-processing chain composed of Fast Fourier Transform, Filter and Short Time Fourier Transform to obtain time-velocity maps or spectrograms allowing the extraction of features from gait and human activities. An encouraging implementation (using SIMD logic) on Jetson Xavier allows us to move to data stream processing. Continuous monitoring of the subject will save lives, minimize injuries, reduce anxiety and prevent post-fall syndrome (PDS).