Similar play estimation using only motions with automatic generation of template space-time map

Plays in sports are described as the motions of players. This paper proposes the method to create a better template for retrieving the plays using only motion compensation vectors in MPEG videos. This work uses space time images for describing the motion in videos. The space time image is a connected sequence of 1-dimensional degenerated descriptions of each motion image between two adjacent frames. This space time image describes a sequence of frames as a 2-dimensional image. This work shows the method to create new better templates from a single template for retrieving a small number of plays in a huge number of frames using both the correlation and the absolute difference. Our method uses the K-means clustering and simple averaging to create new templates. Our experiment records 0.946 as recall, 0.869 as precision and 0.906 as F-measure in 168 plays in 132485 frames.