Expression of cooperative motions in team sports based on the motion data of each player

Abstract

In this paper, we propose a method of generating action description and detecting certain combination plays observed in volleyball games. Motions are represented as motion symbols with lower dimensions, and positions of the ball at each time are obtained. These motion symbols and ball positions are used to detect and describe motions. In detecting combination plays, motion pattern at each time and each players are compared with reference motion patterns selected by hand, and similar motion patterns are detected. In generating motion descriptions, motion symbols and ball positions are used as the initial state of Gated Recurrent Unit, and it learns to generate motion descriptions. This method was tested using motion data obtained from real volleyball match.