Similar Situations Identification for the Game of Soccer

抄録

In this work, we study the repeatability of game situations in different soccer matches. This analysis is aimed to evaluate the possibility of reusing these records as part of a game AI system. Due to a variety of team formations, an appropriate comparison of game situation pairs is a challenging task. Identification of similar situations in the game of soccer can be presented as an evaluation of geometrical similarity of players’ coordinates on the field. Team formations have semantic value, and we show that role-based analysis is essential for successful matching. Obtained results can be applied for the tasks of sports analytics and AI design.