Scholarlink

Background <p>In orthopedics and sports medicine, understanding biomechanics is essential for optimizing performance and preventing injuries. Traditional motion-capture methods are often impractical in dynamic, real-world sports settings. Recent advances in artificial intelligence (AI)-powered video analysis, especially markerless motion tracking, offer promising solutions to these challenges. However, applying these technologies in unconstrained environments like live sports remains complex.</p> Objective <p>This article aims to review state-of-the-art computer vision methods used in sports medicine, focusing on the challenges and solutions of AI-based video analysis in real-world settings. It evaluates the effectiveness of monocular and multi-view systems in analyzing athlete motion and biomechanics during live competition.</p> Methods <p>We review the latest video-based 3D human motion analysis techniques from the past 5&#xa0;years, focusing on challenges such as occlusion, camera calibration, and multi-person tracking in sports environments. We highlight open-sourced algorithms and their applications, including monocular and multi-view approaches for biomechanical assessments.</p> Results <p>Artificial intelligence-based video analysis has shown significant progress, with monocular models achieving reliable results in controlled environments and multi-view systems, improving tracking accuracy in dynamic settings. Despite these advancements, issues like occlusion, synchronization, and limited real-world data still hinder broad application. Data variability and the need for personalized models remain significant challenges.</p> Conclusion <p>While monocular systems excel in controlled environments, multi-view setups are essential for accurate analysis in team sports. Future developments must balance model accuracy with practical implementation in diverse sports contexts. Collaboration between clinicians, engineers, and industry stakeholders will be crucial for advancing AI-powered video analysis in sports medicine.</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

Artificial intelligence for motion analysis

  • 期刊论文

  • 发表日期：

    2025年5月16日

查看全文
  • Attila Schulc,
  • Jakob Ackermann,
  • Adrian Deichsel,
  • Brenda Laky,
  • Lukas N. Münch,
  • Larissa A. Eckl,
  • Dominic T. Mathis,
  • Lena Eggeling,
  • Arasch Wafaisade,
  • Christoph Kittl,
  • Karl F. Schüttler,
  • Daniel Günther,
  • Gergo Merkely

摘要

Background

In orthopedics and sports medicine, understanding biomechanics is essential for optimizing performance and preventing injuries. Traditional motion-capture methods are often impractical in dynamic, real-world sports settings. Recent advances in artificial intelligence (AI)-powered video analysis, especially markerless motion tracking, offer promising solutions to these challenges. However, applying these technologies in unconstrained environments like live sports remains complex.

Objective

This article aims to review state-of-the-art computer vision methods used in sports medicine, focusing on the challenges and solutions of AI-based video analysis in real-world settings. It evaluates the effectiveness of monocular and multi-view systems in analyzing athlete motion and biomechanics during live competition.

Methods

We review the latest video-based 3D human motion analysis techniques from the past 5 years, focusing on challenges such as occlusion, camera calibration, and multi-person tracking in sports environments. We highlight open-sourced algorithms and their applications, including monocular and multi-view approaches for biomechanical assessments.

Results

Artificial intelligence-based video analysis has shown significant progress, with monocular models achieving reliable results in controlled environments and multi-view systems, improving tracking accuracy in dynamic settings. Despite these advancements, issues like occlusion, synchronization, and limited real-world data still hinder broad application. Data variability and the need for personalized models remain significant challenges.

Conclusion

While monocular systems excel in controlled environments, multi-view setups are essential for accurate analysis in team sports. Future developments must balance model accuracy with practical implementation in diverse sports contexts. Collaboration between clinicians, engineers, and industry stakeholders will be crucial for advancing AI-powered video analysis in sports medicine.