Indoor human motion classification by L1-norm subspaces of micro-Doppler signatures

In this paper, we propose a radar-based human motion classification technique for indoor monitoring of patients and elderly. The proposed method employs L1-norm Principal Component Analysis (L1-PCA) derived subspaces of the associated micro-Doppler signatures. L1-PCA is more robust than standard PCA, exhibiting resistance against outliers among the processed data. Results based on real data experiments demonstrate that the proposed method exhibits performance similar to PCA-based indoor human motion classification when the training data are nominal/clean, while it provides enhanced performance when the training micro-Doppler signature datasets are outlier-corrupted.