Are Existing Monocular Computer Vision-Based 3D Motion Capture Approaches Ready for Deployment? A Methodological...

Area of Science:

• Biomechanics
• Sports Science
• Computer Vision

Background:

• Traditional kinematic data collection in alpine skiing, like stereophotogrammetry, is complex and resource-intensive.
• There is a need for more accessible and efficient methods to analyze skiing technique and performance.

Purpose of the Study:

• To evaluate a monocular computer vision (MCV)-based approach for collecting alpine skiing kinematic data.
• To compare the MCV approach's accuracy and precision against the gold standard (stereophotogrammetry).
• To assess the MCV method's readiness for applied research questions in alpine skiing.

Main Methods:

• Developed and trained deep neural networks using image data from a single camera to predict 3D human pose and ski orientation.
• Collected data from six competitive alpine skiers during a field experiment.
• Validated the MCV approach by comparing its kinematic output with stereophotogrammetry measurements.

Main Results:

• The MCV approach achieved a normalized mean per joint position error of 0.08 ± 0.01m.
• High accuracy and precision were observed for joint angles (e.g., knee flexion, hip flexion) and skiing-specific metrics (e.g., center of mass position, lean angle).
• The obtained accuracy and precision magnitudes are deemed acceptable for detecting significant differences in alpine skiing.

Conclusions:

• The monocular computer vision approach demonstrates high potential as a viable, deployment-ready tool for kinematic analysis in alpine skiing.
• This technology offers a more accessible alternative to traditional methods, facilitating applied research and performance analysis.
• The findings support the use of MCV for detailed biomechanical assessments in dynamic sports like alpine skiing.