Scholarlink

<p>This research developed and validated a comprehensive multimodal assessment system integrating artificial intelligence for aerobic gymnastics training evaluation. A 12-week randomized comparative study recruited 600 participants (300 national-level athletes, mean age 22.4 ± 3.2&#xa0;years; and 300 university students, mean age 20.8 ± 2.1&#xa0;years) to establish normative performance databases and validate system capabilities across different skill levels. The integrated framework combined video analysis, motion capture, physiological monitoring, cognitive tracking, and psychological assessment through deep learning architectures. The CNN-RNN stream (video and motion) achieved 96.4% accuracy (95% CI 95.8–97.0%) and the five-modality fusion 93.7%, with 94.1% operational stability and 78&#xa0;ms latency. Analysis revealed critical training parameters including shoulder flexion range of 165.3 ± 12.7° across the cohort, heart rate variability fatigue threshold at 28.7&#xa0;ms RMSSD, and optimal training duration of 40–80&#xa0;min for maintaining peak psychological state. Group training showed clear advantages over individual practice, with social connectedness rising from 2.1 ± 0.28 to 4.2 ± 0.35 (<i>p</i> &lt; 0.001). Strong correlations emerged between heart rate variability (HRV) and positive affect (r = 0.76), establishing physiological biomarkers for psychological readiness. Multimodal analysis revealed interactions between technical performance, physiological condition, and psychological components that single-modal approaches could not capture. This framework has the potential to support personalized training optimization in aerobic gymnastics.</p><p>Trial registration: This AI-driven multimodal system demonstrated high accuracy and feasibility for assessing aerobic gymnastics performance. This trial was registered retrospectively with ISRCTN (ISRCTN14809597) on 06 July 2026.</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

Development and validation of an AI-driven multimodal system for assessing aerobic gymnastics training using video analysis, motion capture, and physiological signals

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月10日

查看全文
  • Mengke Zhu,
  • Mei Gong,
  • Gang Qin,
  • Wei Wang

摘要

This research developed and validated a comprehensive multimodal assessment system integrating artificial intelligence for aerobic gymnastics training evaluation. A 12-week randomized comparative study recruited 600 participants (300 national-level athletes, mean age 22.4 ± 3.2 years; and 300 university students, mean age 20.8 ± 2.1 years) to establish normative performance databases and validate system capabilities across different skill levels. The integrated framework combined video analysis, motion capture, physiological monitoring, cognitive tracking, and psychological assessment through deep learning architectures. The CNN-RNN stream (video and motion) achieved 96.4% accuracy (95% CI 95.8–97.0%) and the five-modality fusion 93.7%, with 94.1% operational stability and 78 ms latency. Analysis revealed critical training parameters including shoulder flexion range of 165.3 ± 12.7° across the cohort, heart rate variability fatigue threshold at 28.7 ms RMSSD, and optimal training duration of 40–80 min for maintaining peak psychological state. Group training showed clear advantages over individual practice, with social connectedness rising from 2.1 ± 0.28 to 4.2 ± 0.35 (p < 0.001). Strong correlations emerged between heart rate variability (HRV) and positive affect (r = 0.76), establishing physiological biomarkers for psychological readiness. Multimodal analysis revealed interactions between technical performance, physiological condition, and psychological components that single-modal approaches could not capture. This framework has the potential to support personalized training optimization in aerobic gymnastics.

Trial registration: This AI-driven multimodal system demonstrated high accuracy and feasibility for assessing aerobic gymnastics performance. This trial was registered retrospectively with ISRCTN (ISRCTN14809597) on 06 July 2026.