Statistical Analysis of Piano Performance Hand Motion Data and Deep Learning Model for Loss Recovery

In one sentence

Recovers up to 49% missing data in OptiTrack piano hand motion capture using TimesNet-based outlier detection and deep-learning imputation.

Key facts

• ~2 hours of motion capture from one professional pianist
• Up to 49% missing values during capture
• MCAR 20% test: MAE 0.2482, RMSE 0.4948

When this paper is relevant

• You impute missing motion-capture data with deep learning
• You want statistics on missing-data patterns in piano hand motion

Related

• Automated gap-filling algorithm for kinetic data of finger movement: Pianist hand motion cleaning using spatio-temporal imputation (2023)
• SKY-Piano: A Multimodal Piano Performance Dataset (2026)

Cite

배준형, 권태균, 김기락, 박다원, 최윤석, 허현, 박재범, 박종화, 남주한 (2025). Statistical Analysis of Piano Performance Hand Motion Data and Deep Learning Model for Loss Recovery. 68th Conference of the Korean Society for Music Perception and Cognition, Feb. 20, 2025.

@inproceedings{Bae2025hand,
  title = {Statistical Analysis of Piano Performance Hand Motion Data and Deep Learning Model for Loss Recovery},
  author = {배준형 and 권태균 and 김기락 and 박다원 and 최윤석 and 허현 and 박재범 and 박종화 and 남주한},
  booktitle = {68th Conference of the Korean Society for Music Perception and Cognition, Feb. 20, 2025},
  year = {2025}
}

Machine-readable copy: /papers/hand-motion-loss-recovery-2025.md