Інтелектуальний аналіз складності гімнастичних комбінацій засобами комп’ютерного зору

Abstract

This study presents an intelligent system for automated gymnastics routine difficulty assessment using computer vision. The pipeline employs MediaPipe for skeletal pose extraction, a finite state machine for temporal segmentation, and a bidirectional GRU network for element classification (77% accuracy). Recognized elements are mapped to predefined difficulty coefficients and group requirements to compute the D score. The approach enables consistent, repeatable evaluation and supports sports officiating and performance analysis.