Fall detection with multi-domain features by a portable FMCW radar

Abstract

Fall detection is important for senior care. In order to classify fall and other fall-similar daily motions, a novel dynamic range-Doppler trajectory (DRDT) method based on a frequency-modulated continuous-wave (FMCW) radar system is proposed. Multi-domain features including temporal changes of range, Doppler, radar cross-section (RCS) and dispersion are extracted from echo signals for a subspace K-Nearest Neighbor (KNN) machine learning classifier. Extensive experiments demonstrated its feasibility and an average accuracy of 95.5% was achieved in recognizing six typical fall-similar motions.