Pose-derived biomechanical feature profiling and proxy pattern classification of basketball player movements

Area of Science:

• Biomechanics
• Sports Science
• Computer Vision

Background:

• Markerless pose estimation offers a practical alternative to lab-based motion capture for sports analysis.
• Interpreting pose-derived biomechanics requires caution due to potential lack of ground-truth action labels and validated outcomes.

Purpose of the Study:

• To develop a computational pipeline for biomechanical feature profiling of basketball player movements.
• To utilize annotated gameplay pose data for deriving and analyzing movement-related variables.

Main Methods:

• Processed frame-level keypoint annotations from the TrackID3x3 dataset.
• Derived biomechanical variables (e.g., joint angles, segment distances, asymmetry, trunk orientation).
• Employed a sliding-window approach for temporal sequence analysis and rule-based proxy groupings for evaluation.

Main Results:

• Generated a structured dataset of 37,134 frame-player observations.
• Demonstrated measurable variations in joint alignment, limb symmetry, and trunk orientation.
• Classification results indicate proxy-label separability, not validated action prediction.

Conclusions:

• The developed pipeline provides a method for descriptive biomechanical analysis of basketball players using pose data.
• Results highlight the need for action-level annotations and independent biomechanical validation for robust interpretation.
• Future research should integrate these elements to enhance the accuracy and applicability of pose-derived biomechanical analysis.