EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Explore the application of multi-scale feature fusion of sequences in the estimation of dance movement postures

Yafen Chen, Nikolai Tsvetkov

Data and Metadata · 16.04.2026 · 5 · 810

DOI: 10.56294/dm2026810 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Data and Metadata, online erschienen am 16.04.2026. Verfasst von Yafen Chen und 1 weiteren Autor:innen. Die Arbeit nennt 21 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Introduction

Dance movement posture estimation plays a crucial role in performance analysis and training support. Accurately modelling human motion presents challenges due to rapid movements, intricate joint articulations, and varied postures. Existing methods often rely on single-scale features or frame-wise models, which fail to capture fine joint details and long-term temporal patterns, leading to unreliable posture predictions.

Objective

This research proposes a multi-scale feature fusion model, the Snow Leopard-driven Temporal Convolutional Neural Network (SL-driven TempConvoNet) for precise dance posture estimation.

Method

The dataset used consists of 2000 RGB video recordings, covering multiple participants performing diverse dance sequences. Data pre-processing involves frame resizing, z-score normalization, cropping around the dancer to reduce background noise, and Gaussian filtering to standardize sequence lengths. Feature extraction is performed using Histogram of Oriented Gradients (HOG) to capture local posture and edge information, while SL-driven TempConvoNet models temporal dependencies across frames. The TempConvoNet lies in its adaptive multi-scale temporal modeling, fusion of spatial HOG features with temporal representations, and SLOA is to efficiently adjust parameters and fine-tune the model’s performance.

Results

Experimental results demonstrate superior performance with 98,9% accuracy, 98,8% precision, 92,5 % recall, 98,8 % F1-score, and 98,9 % sensitivity[user2.1], significantly outperforming conventional methods, using Python tools for model evaluation and performance metrics calculation.

Conclusion

This results in improved motion continuity, posture accuracy, and computational efficiency, offering a reliable and efficient solution for dance analysis, intelligent training, and interactive performance systems.

Zitieren

APA

Chen, Y., & Tsvetkov, N. (2026). Explore the application of multi-scale feature fusion of sequences in the estimation of dance movement postures. Data and Metadata, 5, 810. https://doi.org/10.56294/dm2026810

Vancouver

Chen Y, Tsvetkov N. Explore the application of multi-scale feature fusion of sequences in the estimation of dance movement postures. Data and Metadata. 2026;5:810. doi:10.56294/dm2026810

BibTeX
@article{Chen2026,
  author = {Chen, Yafen and Tsvetkov, Nikolai},
  title = {Explore the application of multi-scale feature fusion of sequences in the estimation of dance movement postures},
  journal = {Data and Metadata},
  year = {2026},
  volume = {5},
  pages = {810},
  doi = {10.56294/dm2026810}
}
Kostenlos starten →