Subspace Classification of Human Gait Using Radar Micro-Doppler Signatures

Radar-based monitoring of human gait has become of increased interest with applications to security, sports biomechanics, and assisted living. Radar sensing offers contactless monitoring of human gait. It protects privacy and preserves a person's right to anonymity. Considering normal, pathological and assisted gait, we demonstrate the effectiveness of radar in discriminating different walking styles. By use of unsupervised feature extraction methods utilizing principal component analysis, we examine five gait classes using two different joint-variable signal representations, i.e., the spectrogram and the cadence-velocity diagram. Results obtained with experimental K-band radar data show that the choice of signal domain and adequate pre-processing are crucial for achieving high classification rates for all gait classes.