Human-Centered AI
2023-04-01 – 2025-12-31 · Completed
Institute of Information & Communications Technology Planning & Evaluation (IITP)
Research
- Combined temporal, frequency and spatial rPPG features, including CSP spatial features, for physiological-signal estimation.
- Used speech–text contrastive learning for conversational emotion recognition.
- Contributed gaze-based engagement and full-body anomaly models to joint LucyCare kiosk demonstrations.
Results
Joint development results reported in the 2026 final evaluation.
| Task | Evaluation result |
|---|---|
| Heart rate from facial video | MAE 2.53 bpm |
| HRV from facial video | MAE 24.58 ms |
| Speech–text emotion recognition | 85.9% accuracy · IEMOCAP/MELD model evaluation |
| Engagement prediction | 70.4% · project evaluation criteria |
| Full-body anomaly detection | 76.0% accuracy · project evaluation criteria |
Speech–text emotion accuracy increased from 83.8% in year two to 85.9% in year three, a gain of 2.1 percentage points.