Research Article
A Multitask Sign Language Recognition System Using Commodity Wi-Fi
Table 3
Recognition accuracy of Wi-SignFi and other methods.
| Methods | Modality | Home276 (%) | Lab276 (%) | Lab + Home276 (%) | Lab150 (%) | Mean (%) | Rank |
| SignFi [3] (2018) | A + P | 98.91 | 98.01 | 94.81 | 86.66 | 94.60 | 6 | HOS-Re [33] (2019) | — | 98.26 | 97.84 | 96.34 | 96.23 | 97.17 | 5 | CNN [31] (2020) | A + P | 99.64 | 99.855 | 99.73 | 93.84 | 98.27 | 3 | Dual-output two-stream with ResNet50 [4] (2020) | A + P + Difference | 99.13 | 96.79 | 97.08 | 95.88 | 97.22 | 4 | LSTM [32] (2021) | A | 99.50 | 99.80 | 99.40 | 78 | 94.18 | 7 | Wi-SignFi (224 × 224 × 3) | A + P | 99.75 | 99.80 | 99.69 | 96.41 | 98.91 | 1 | Wi-SignFi (200 × 180) | A + P | 99.71 | 99.67 | 99.40 | 96.79 | 98.89 | 2 |
|
|
The bold values given in Table 3 represent the best action recognition results for each of the four datasets.
|