Development of a MediaPipe-based framework for biomechanical quantification of table tennis forehand strokes

This study aimed to quantify kinematic relationships across body segments during forehand strokes to provide interpretable metrics for single-camera based lightweight table tennis diagnostics.

We analyzed 34 female players (aged 9.1-21.7 years) from provincial teams, recording a total of 340 strokes (10 per player). An SVM model was used to predict ball speed, after which 320 strokes (8-10 per player) were retained by removing outliers in ball speed. From MediaPipe position time series, we calculated velocity, angle and angular velocity time series, and extracted kinematic parameters from these time series, including range, mean/peak/impact values. Within-subject correlation coefficients (r _ws) were calculated to identify key biomechanical parameters that contribute to the ball speed, while between-subject correlation coefficients (r _bs) were used to detect the relationship between age/height and ball speed.

Ball speed increased with greater playing-side arm linear movement at the shoulder (r _ws = 0.51 to 0.63), elbow (r _ws = 0.63 to 0.70) and wrist (r _ws = 0.50 to 0.60), as well as with enhanced rotational motion at the playing-side upper arm (r _ws = 0.65 to 0.71), shoulder line (r _ws = 0.54 to 0.57), and hip line (r _ws = 0.51 to 0.59). Conversely, ball speed decreased with excessive contralateral shoulder horizontal flexion/extension (r _ws = -0.44 to -0.62) and playing-side elbow flexion-extension (r _ws = -0.35). At the population-level, ball speed increases with age before 14.3 years (r _bs = 0.68) but plateaus thereafter (r _bs = 0.17).

This MediaPipe-based framework demonstrates potential for efficient biomechanical analysis in table tennis, providing a promising foundation for lightweight real-time analysis solutions.