Abstract
Recently, we have been developing a system that automatically acquires tactics and strategies
of the table tennis match from broadcast video. In this system, the input and output data are constructed
by automatically extracting the ball position and the player position from the broadcast video. During the
match, an algorithm removes noise and estimates the ball and player position. In this paper, we introduce
the motion tracking system. In the motion tracking system, the ball trajectory and player position are
automatically extracted from the 30fps broadcast video and converted into two-dimensional coordinates.
The ball trajectory is estimated by preprocessing with the white blog extraction process and RGB extraction
process, and then the player’s skeleton position is estimated by CenterNet of deep learning (DNN). The
position of the ball hidden at the body is estimated using the Kalman filter and the bicubic interpolation
method. Finally, we discuss the future image of this system which acquires the table tennis strategy and
makes the strategy visible to directors and coaches using the if-then rule by fuzzy ensemble learning.
