A 3D motion capture data-driven volleyball action optimization model

This study developed a 3D motion capture data-driven model for volleyball action optimization. Twelve synchronized cameras and two force plates were used to collect data from 24 volleyball players (12 professionals and 12 amateurs) performing four key actions. The model integrated kinematic and kinetic features through a hierarchical fusion framework. It achieved a classification accuracy of 92.3% (F1 - score = 0.91), outperforming traditional biomechanical rules - based analysis, SVM, and vanilla LSTM networks. The root mean squared error for joint angles was 3.1°. For overhand serves, shoulder internal rotation increased by 6.6° (p