Real-Time Spin Estimation of Ping-Pong Ball Using Its Natural Brand

Predicting the trajectory of flying objects with spin is a challenge but an essential task in many fields, especially in military and sports. Robots playing ping-pong is a very good platform to validate the trajectory prediction method. Various vision systems have been proposed, but only position information was used in most cases, which limits their capability to predict the trajectory of the spinning ball. Based on the fact that a spinning ball's motion can be separated into translation movement and spinning with respect to the ball's center, this paper proposes a novel vision system that can provide both the position and the spin information of a flying ball in a real-time mode with high accuracy. With a frame difference-based recognition method, the natural brand of a ball can be recognized under normal illumination conditions. Then the 3-D pose of the ball can be restored in ball coordinates. With the observation and analysis that the axis and angular speed of spin do not change during flying, the spin state can be estimated using a weighted-random sample consensus-based plane fitting method. Combining both position and spin information in a force-based dynamic model, accurate trajectory prediction can be achieved via an extended Kalman filter. Experimental results show the effectiveness and precision of the proposed method.