抄録
In this paper, we proposed a new method to identify trajectories and bound positions of a table tennis ball on the table. Our method used audio and visual signals obtained by a microphone and camera placed on the ceiling of the gymnasium. The ball was extracted from the images obtained by the camera. Our method used the fact that the shape of the ball was almost circular in the image, and we developed a rapid method to extract the ball by using image moment. Then the trajectory of the ball on the table was estimated by connecting the successive ball positions. The bound time was estimated by band-pass filtering in which pass-frequency corresponds to the resonant frequency of the ball diameter. The bound position was calculated by interpolating the position at the bound time. Our method was simple, and the computational time was less than the video frame interval. We conducted an experiment to examine the ball bound time on the table from a practice game. The experimental result showed that 98.9% of bounds can be extracted with a precision of 94.4%. The improvement of sound detection accuracy remained to be solved, in order to reduce the miss and over-detection of the bound. Furthermore, we made a manuscript for training regarding the bound position and ball speed on the table based on the obtained data. We could make document contents quickly and objectively.
