Table 2 Classification performance of the proposed DSwarm-Net model on three benchmark HAR datasets. Acc: Accuracy, F1: F1-Score.
From: A union of deep learning and swarm-based optimization for 3D human action recognition
Encoding Used | UTD MHAD | NTU RGB+D 60 | HDM05 | |||||
|---|---|---|---|---|---|---|---|---|
Cross-subject | Cross-view | |||||||
Acc (%) | F1 (%) | Acc (%) | F1 (%) | Acc (%) | F1 (%) | Acc (%) | F1 (%) | |
Distance encoded | 95.62 | 96 | 84.49 | 85 | 87.24 | 88 | 88.45 | 89 |
Angle encoded | 96.81 | 97 | 84.92 | 85 | 88.39 | 89 | 89.17 | 89 |
Distance velocity encoded | 90.23 | 91 | 83.44 | 84 | 87.15 | 88 | 88.34 | 89 |
Angle velocity encoded | 94.55 | 95 | 84.63 | 85 | 88.15 | 89 | 89.09 | 90 |
Compact distance encoded | 97.56 | 98 | 84.81 | 85 | 88.92 | 89 | 89.46 | 90 |
Compact angle encoded | 97.97 | 98 | 84.97 | 85 | 88.66 | 88 | 89.88 | 90 |
DSwarm-Net | 98.13 | 98 | 85.45 | 86 | 89.98 | 90 | 90.67 | 92 |
