Table 1.
Comparison between various health-monitoring methods.
Fig 1.
Overview of the FMCW radar: (a) The FMCW radar block diagram, (b) FMCW sawtooth waveform.
Fig 2.
FMCW processing flow from the IF signal.
Fig 3.
The spectrogram of walking and walking to fall actions in the noise-added dataset with different SNR levels.
Fig 4.
The diagram of the proposed approach.
Fig 5.
Range profile and FFT of a specific chirp.
Fig 6.
The FFT representation at one frame in the STFT processing and the denoising result at -5-dB SNR.
Fig 7.
The overall architecture of the proposed CRCNN.
Fig 8.
Performance of CRCNN with different numbers of filter channels.
Table 2.
Average classification accuracy.
Table 3.
Compare results with different size filters in the im-res block.
Fig 9.
The average classification accuracy of CRCNN depends on SNR with varied numbers of C-R connections.
Table 4.
Comparison results of CRCNN with state-of-the-art DCNN models.
Fig 10.
The spectrogram with various selected range-bin intervals at 5-dB SNR.
Table 5.
Entropy information with different selected range-bin interval.
Fig 11.
The spectrogram images of walking and walking to fall actions at -5-dB SNR with different cut-threshold values.
Fig 10a–10d show the walking action. Fig 10e–10h show the walking to fall action.
Table 6.
Average classification accuracy with varied cut-threshold values.
Fig 12.
The spectrograms of different denoising methods for walking action.
Table 7.
Comparison results of the proposed approach with existing denoising methods.
Fig 13.
Performance comparison of different classifiers.
Fig 14.
Three metrics: Precision, recall, and F1-socer of different classifiers.
Table 8.
Performance metrics of different classifiers at various SNR levels on two datasets.
Table 9.
Classification accuracy with various distances.
Fig 15.
Classification accuracy with various aspect angles.