Główne pojęcia
This research proposes a novel method for generalizable indoor human activity recognition using through-the-wall radar, addressing the challenge of varying human physiques by employing micro-Doppler corner point cloud representation and dynamic graph learning for robust and accurate activity classification across different individuals.
Statystyki
The proposed method achieves a validation accuracy of 94.63% on simulated data and 93.38% on measured data.
The test accuracies on simulated data sets are 94.5%, 90.0%, 87.0%, and 78.5% corresponding to 1.8, 1.7, 1.6, and 1.5 meters tall testers.
The test accuracies on measured data sets are 93.0%, 86.0%, and 80.2% corresponding to 1.8, 1.7, and 1.6 meters tall testers.
The validation accuracy of the proposed method decreases by no more than 15% on simulated data and 10% on measured data as the SNR decreases to 12 dB.