|
| References | Data set | Method | Number and type of radar | Recognition accuracy |
|
| [16] | 20 volunteers
2,880 radar signals each lasting 10 seconds | A deep network constructed from several individually sparse AEs
The input data are μ D signals | Using a single radar
A continuous wave Doppler radar, the ST200 system by RFbeam | The recognition accuracy of single person on random routes can reach 96.20%. |
|
| [17] | 8 volunteers
Total 40,000 frames | Multichannel 3D convolutional neural network
Gait point cloud data as input | Using a single radar
IWR1443BOOST | The recognition accuracy of single person on random routes can reach 93% |
|
| [18] | 10 volunteers
Total 1500 samples
2.3 seconds per sample | Input the gait spectrum map into the convolutional neural network | Using a single radar
AWR1443 | The recognition accuracy of single person on fixed and random routes can reach 91% and 90% |
|
| [19] | 4 volunteers
100000 frames of data | Input point cloud data into multibranch CNN networks | Using a single radar
CAL60S244 IBM AiP | The recognition accuracy of single person on random routes can reach 85.8% |
|
| [20] | 6 volunteers
Three environments
Total 75 minutes of data | Input point cloud data into spatio-temporal graph convolutional network | Using a single radar
IWR6843ISK | The recognition accuracy of single person and two persons on random routes can reach 68.88% and 75.51% |
|
| [14] | 95 volunteers
Total 30 hours of data | Input the five attributes of the point cloud into the CNN network | Using two radars
IWR6843 and IWR1443 | The recognition accuracy of single person and two persons on fixed route can reach 90% and 86% |
|
| [15] | 5 volunteers
150 minutes of gait data | Input the micro-Doppler signature into the DCNN model | Using a single radar
An FMCW radar device produced by industrial radar systems GmbH | The method achieved an error rate of 24.70% on the validation set and an error rate of 21.54% on the test set |
|
| [13] | 12 volunteers
Total 120 minutes of data | Each frame of data is flattened into a 16000 dimensional vector and fed into the bidirectional LSTM | Using a single radar
IWR1443 | The recognition accuracy of single person on random routes can reach 89% |
|
