Visual Measurement And Control For Table Tennis Robot

The robot should track the fast moving target and predict the future tra-jectory of the ball beforehand when the robot plays table tennis games. Also, itshould strike the ball back in proper velocity, position, posture and right strikingtime. These mean that the robot should interact with the environment contin-uously. So the research of the table tennis robot contains visual perception andintelligent control and so on. By the study of the robot, a series of perceptionaland control problem can be investigated.In this thesis, the visual measurement and control for the table tennis robotare investigated. The main contributions of this thesis include following issues:1 A distributed parallel processing High-Speed Vision System based on smartcamera is presented. Compared to traditional vision system based on im-age acquisition card, this vision system makes good use of the powerfulimage processing performance. It transfers the serial mode to parallel anddistributes the computation to each camera to improve the real-time per-formance.2 A set of e?ectiveness table tennis tracking algorithms to track the fast mov-ing ball is proposed. The algorithms integrate adjacent frame di?erence,dynamic window, run length code together. They have good real-time per-formance. Also, the GSP algorithm is proposed to compute the center ofthe ball when the result of the adjacent frame di?erence may be a crescentto improve the accuracy of the 3D position of the ball.3 The analytic solution of the ?ying model is deduced which is helpful toreduce the time-consuming of the prediction algorithm. An e?ective reboundmodel is proposed. In the horizontal direction a linear model is proposedto represent the velocity change before and after rebound on the table.In vertical direction, the rebound coe?cient is employed to compute thevelocity change of the ball. Three phases strike method is proposed withstraight line shot in particular cases.4 Because the ?ying model is very sensitive to the initial velocity, a negativefeedback method based on position information is proposed.5 A three-stage trajectory planning method is proposed. The 5-DOF tabletennis robot can rally with human for 50 rounds base on this method. A3D simulation vision system is designed to redisplay the experimental sceneand adjust the model parameters. Finally, the research results are summarized and future work is discussed.