Author Affiliations
Tracking Guidance Teaching and Research Section, Air Defense and Missile Defense College, Air Force Engineering University, Xi’an, Shaanxi 710051, Chinashow less
Fig. 1. Flow chart of echo signal processing
Fig. 2. Comparison of time domain and frequency domain among different gestures. (a) Gesture 1; (b) gesture 2; (c) gesture 3; (d) gesture 4
Fig. 3. Flowchart of system working
Fig. 4. Feature map clustering analysis
Fig. 5. Clustering results for different K values. (a) Original image; (b) K=2; (c) K =4; (d) K =8
Fig. 6. Visualization of clustering results
Fig. 7. Random forest visualization
Fig. 8. Contribution rate of some features
Fig. 9. SNR of maps with different channel numbers
Number of accumulated echo pulses | 48 | 96 | 128 | 256 |
---|
Average accuracy of classification /% | 67.23 | 81.46 | 98.67 | 99.97 |
|
Table 1. Comparison of recognition accuracy for different number of accumulated echo pulses
Type of feature map | Average accuracy of classification /% |
---|
RGB map | 98.67 | R channel map | 77.98 | G channel map | 81.25 | B channel map | 78.25 | Gray-scale map | 83.54 |
|
Table 2. Comparison of accuracy for different feature maps
Type of input | CART number | Maximum number of features | Average accuracy of classification /% |
---|
RDM | 50 | 6 | 79.26 | 12 | 85.32 | 20 | 86.17 | 100 | 6 | 87.34 | 12 | 98.67 | 20 | 99.12 | 200 | 6 | 86.54 | 12 | 97.12 | 20 | 98.31 |
|
Table 3. Results of classification for different parameter settings of random forest
Parameter | Setting |
---|
Number of transmitting antennas | 1 | Number of receiving antennas | 1 | Number of integrated radar pulses | 128 | RDM frame duration /ms | 14 | Number of sampling points | 8000 | Type of input | RDM | Size of input | 512×512×3 | CART number | 100 | Maximum number of features | 12 | Number of gesture classes | 6 |
|
Table 4. Setting of parameters in gesture recognition algorithm
Algorithm | Training time /h | Recognition speed /(frame·s-1) | Accuracy /% |
---|
ShuffleNet V2 | 4.46 | 41 | 90.64 | Mobilenet V2 | 4.52 | 42 | 91.53 | VGG 16 | 4.21 | 46 | 86.63 | Algorithm proposed in this article | 2.61 | 41 | 98.93 |
|
Table 5. Comparison of proposed algorithm with other algorithms