Accessory Based Classification of Football Players Using Attention UNet

Accurate soccer player classification is essential in the realm of sports video analytics, allowing for comprehensive evaluations of player performance and potential. In this study, we propose the adoption of an Attention UNet model for precise player categorization based on their distinct accessories. The innovative Attention UNet architecture, equipped with attention gates, excels at capturing and enhancing pertinent features, ensuring highly accurate classification. We conducted evaluations using a carefully curated custom dataset comprising 500 images depicting goalkeepers, referees, and captains. Performance metrics including Mean Squared Error (MSE) loss, Intersection over Union (IoU), dice coefficient, precision, and recall were rigorously applied. In stark contrast to existing soccer player classification methods, our approach demonstrates substantial performance enhancements, achieving an impressive accuracy rate of 98.93%. This research fills a critical need for precise player classification, offering invaluable insights for player tracking, performance analysis, and data-driven decision-making in the domain of soccer analytics. By providing this innovative framework, we aim to elevate the capabilities of sports video analytics, benefiting researchers, analysts, and sports enthusiasts who look for advanced tools for player assessment and talent identification.