METHODOLOGY FOR THE USE OF NEURAL NETWORKS TO ASSESS THE QUALITY OF TRAINING IN SPORTS

Abstract

The article proposes a methodology for analyzing sports movements using neural networks, integrating the stages of pose estimation, construction of a digital movement skeleton, and comparison
with reference trajectories. The approach is based on the use of modern neural network architectures
(HRNet, OpenPose, ViTPose, PoseFormer, KASportsFormer) and specialized sports datasets to ensure high accuracy in recognizing an athlete’s pose and interpreting the dynamic phases of performing sports exercises. A methodology has been developed focused on the formulation and application
of requirements for an information system with the prospect of its use in sports education and coaching practice.