Real-Time Visual Measurement With Opponent Hitting Behavior for Table Tennis Robot

In this paper, a visual-based method is proposed for the table tennis robot to estimate the hitting point from the opponent, which can provide a better understanding of the opponent, and thus improve the responsiveness of the table tennis robot. As an essential manifestation of opponent hitting behavior, the information of hitting point includes not only the motion status variation of the ball before and after being hit, but also the racket pose at the hitting moment. To solve this problem, the trajectories of the ball before and after being hit are first predicted based on visual measurement and the motion model of the ball. The racket trajectory of the opponent is then derived by a self-adaptive threshold selection scheme and a multifilter. Considering that the ball and racket trajectories are not absolutely precise, an optimized solution is proposed to compute the hitting point from the opponent. To the best of our knowledge, the proposed approach is the first one that could achieve a fast estimation of the hitting point from the opponent with a satisfactory resolution. The effectiveness of the proposed approach is verified by experiments.