the trajectory prediction of a ping-pong ball for robots playingto demonstrate a feasible solution to such a problem.Ping-pong robots have become a hot research topic in lastdecades for it is an ideal platform for real-time hand–eyecoordination experiments. A vision system that can detect,locate the moving ball in 3-D coordinates, and predict itsfollowing trajectory is essential for successful playing. Variousvision systems have been developed. Earlier, monocular visionsystems [6], [7] were widely employed. Because only onecamera is used, shadow becomes very important for the ball’s3-D localization, and thus, they have restricted requirementson the illumination conditions and the ball’s flight velocityand area. Recently, stereo vision systems [8]–[11] andmultivision systems [12], [13] are used in most ping-pongrobot systems, which can provide more robust and accurateresults of the ball’s location. However, by now, almost allof the vision systems for ping-pong robots provide only theposition of the ball. In this paper, they are called positionvision systems.Limited by the ob