Towards autonomous evaluation of horse-rider performance in equestrian sports utilizing on-field data and artificial intelligence

Abstract

This thesis presents a comprehensive methodology for the quantitative evaluation of horse-rider performance, based on the dimensionality reduction of both on-field dressage trajectories and body kinematics data. The proposed approach leverages Singular Value Decomposition (SVD) and Uniform Manifold Approximation and Projection (UMAP) to extract low-dimensional representations capturing the essential features of athlete behaviour, used in their evaluation.

Two evaluation strategies were implemented: a distance-based analysis, quantifying athlete consistency and similarity to expert performance, and a machine learning-based expertise prediction. The distance-based analysis showed mixed results, with dressage trajectories producing more pronounced class separation and ranking consistency, while body kinematics exhibited class separation directly related to gait difficulty. In contrast, the approach based on machine learning achieved high prediction accuracies on both training and testing datasets.

Although limited by the small number of athletes, the results highlight the potential of the proposed methodology to identify meaningful features without relying on predetermined parameters or expert knowledge, offering a fully data-driven alternative for athlete evaluation.

Overall, this work demonstrates the feasibility of using dimensionality reduction and machine learning to assess horse-rider performance, providing a foundation for future studies with larger datasets and enabling the extension of this approach to other sports for quantitative performance assessment.