Predictive Sports Strategy Approach Using YOLO and YOLO-NAS in Performance Sports

The usages of You Only Look Once (YOLO) and You Only Look Once Neural Architecture Search (YOLO-NAS) models in digital tracking and strategy extraction from videos are multi-fold. Firstly, these models aim to accurately detect and track players and objects in real time, providing comprehensive data on their movements and interactions. Secondly, they seek to extract strategic elements and patterns from the tracked data to aid in performance analysis and strategy development. Additionally, YOLO-NAS specifically aims to optimize the network architecture for improved tracking and strategy extraction. The goal of this study was to analyse the feasibility of ML-based strategy extraction, and secondly to validate if - in terms of retrospective-type learning methods - athletes and coaches can leverage historical video data to identify trends, analyze past performances, and develop targeted predictive, AI-assisted training programs or adaptive strategies for future competitions. These outcomes will facilitate a novel data-driven approach to athletes and team development, enhancing performance and strategic decision-making in sports.