Research Article
3D Skeletal Human Action Recognition Using a CNN Fusion Model
Table 3
Comparisons with other skeletal action recognition methods on the NTU RGB + D Dataset.
| | Approach | CS | CV |
| | Lie group [21] (2014) | 50.1 | 52.8 | | Part-aware LSTM [12] (2016) | 62.9 | 70.3 | | ST-LSTM + Trust Gate [30] (2016) | 69.2 | 77.7 | | STA-LSTM [31] (2017) | 73.4 | 81.2 | | GCA-LSTM [32] (2017) | 74.4 | 82.8 | | VA-LSTM [33] (2017) | 79.4 | 87.6 | | SR-TSL [33] (2017) | 84.8 | 92.4 | | HCN [34] (2018) | 86.5 | 91.1 | | 2s-AGCN [35] (2019) | 88.5 | 95.1 | | AGC-LSTM [36] (2019) | 89.2 | 95.0 | | DGNN [37] (2019) | 89.9 | 96.1 | | NAS-GCN [38] (2020) | 89.4 | 95.7 | | Proposed (spatial) | 88.3 | 93.6 | | Proposed (temporal) | 83.1 | 86.8 | | Proposed (fusion) | 90.0 | 95.2 |
|
|
