Scholarlink

<p>This study presents a real-time interactive system using pose-based recognition to support learning of Dunhuang dance, an important intangible cultural heritage. A dataset of approximately 1230 annotated images was constructed, capturing correct and incorrect executions of five representative poses. Using MediaPipe for skeletal keypoint extraction, eight machine learning algorithms were compared; Random Forest achieved 98.5% accuracy and 99.4% recall in correctness classification and over 97% accuracy in multi-class pose recognition. The system operates at 20–25 fps with latency below 100 ms, integrating tolerant real-time feedback with Dunhuang-style artefact generation to support both embodied learning and cultural mediation. Results demonstrate that lightweight models can effectively support embodied interaction in data-scarce cultural contexts. Future work will explore temporal modelling and broader deployment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Pose-based embodied interaction for digital Dunhuang dance heritage

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月9日

查看全文
  • Qinyu Fu,
  • Yixuan Zhou,
  • Xiaosheng Ding

摘要

This study presents a real-time interactive system using pose-based recognition to support learning of Dunhuang dance, an important intangible cultural heritage. A dataset of approximately 1230 annotated images was constructed, capturing correct and incorrect executions of five representative poses. Using MediaPipe for skeletal keypoint extraction, eight machine learning algorithms were compared; Random Forest achieved 98.5% accuracy and 99.4% recall in correctness classification and over 97% accuracy in multi-class pose recognition. The system operates at 20–25 fps with latency below 100 ms, integrating tolerant real-time feedback with Dunhuang-style artefact generation to support both embodied learning and cultural mediation. Results demonstrate that lightweight models can effectively support embodied interaction in data-scarce cultural contexts. Future work will explore temporal modelling and broader deployment.