TABLE 3. Baduanjin Action Recognition Results From the Two Models with Different Feature Extraction Methods.
| No. | Pose name | Manual feature extraction | CNN feature extraction |
|---|---|---|---|
| 1 | Propping up the heavens | 64.29 % | 92.86 % |
| 2 | Drawing the bow | 78.57 % | 92.86 % |
| 3 | Separating heaven & earth | 57.14 % | 100.00 % |
| 4 | The wise owl gazes backward | 64.29 % | 100.00 % |
| 5 | The big bear turns from side to side | 57.14 % | 100.00 % |
| 6 | Touching the toes and bending backwards | 64.29 % | 85.71 % |
| 7 | Punching with angry gaze | 50.00 % | 100.00 % |
| 8 | Bouncing on the toes | 92.86 % | 100.00 % |
| Total accuracy | 66.07% | 96.43% |