A digital-humanities project used Kinect skeletal data from Yao Long Drum Dance performers, recovered occluded joints and applied a 3D CNN, reporting classification accuracy up to 96.45%.
Key findings
- The system reported classification accuracy up to 96.45% and proposed a workflow spanning capture, occlusion recovery and digital representation of dance movements.
Why this matters globally
The technique may support archiving and teaching complex performance heritage, while raising questions about control of cultural data and meaning lost when movement becomes coordinates.
Thai researcher contribution
A Krirk University-affiliated researcher contributed computer-vision methods to cultural-heritage research in China.
Limitations to consider
The abstract omits performer and sequence counts, class balance, train-test splitting, baselines and external validation, preventing assessment of leakage and generalization. Consent, community rights and governance are also absent.