Automating Video-Based Two-Dimensional Motion Analysis in Sport? Implications for Gait Event Detection, Pose...

Area of Science:

• Sports Biomechanics
• Computer Vision in Sports
• Movement Analysis

Background:

• Two-dimensional (2D) video analysis is crucial for sports training and performance evaluation.
• Manual annotation of joint centers and kinematic parameters is time-consuming.
• Automated methods using AI and computer vision offer potential efficiency gains.

Purpose of the Study:

• To systematically analyze the automation of 2D video analysis workflows.
• To investigate the applicability of 3D event detection algorithms to 2D video data.
• To assess the agreement of 2D keypoints with 3D motion capture and the impact of event detection offsets.

Main Methods:

• Utilized repeated measures limits of agreement to compare markerless and marker-based motion capture.
• Evaluated a threshold-based event detection algorithm on 2D video at various sampling rates.
• Compared 2D keypoints from pose estimation with projected 3D marker trajectories.

Main Results:

• A minimum video sampling rate of 100 Hz is essential for accurate key event detection.
• 3D marker trajectory-based algorithms showed limited applicability for 2D video analysis.
• Event misidentification (e.g., 20ms touchdown error) led to significant joint angle differences (up to 20°).

Conclusions:

• Existing algorithms for 3D motion analysis are not directly transferable to 2D video.
• Accurate key event detection is critical for reliable automated 2D video analysis.
• Development of novel, de novo algorithms is necessary to enhance 2D video analysis pipelines.