Human-Centered AI
2017-03-01 – 2020-02-29 · Completed
National Research Foundation of Korea (NRF)
Results
- Classified sign gestures with CNNs using composite images of nine video frames.
- Designed gesture-region extraction and similarity networks for distance changes and consecutive gestures.
- Developed semi-automatic training-data generation across people, clothing and backgrounds.
Criteria reported as met in the 2020 completion report.
| Metric | Criterion met |
|---|---|
| Recognition distance | 5 m |
| Environments | At least 27 |
| Recognition accuracy | 85% |