The Design Of Machine Vision For Ping-pong Robot Based On PC

The paper introduces a design of machine vision based on a single-eye robot. The design is consisted of four parts: grabbing driver of the camera, recognition algorithm and obtaining the positions of Ping-pong ball and its shadow, getting 3-D coordinate of ball, and trace prediction of the flying ball. Grabbing driver of the camera includes image format transformation and time stamp making. The information of the shadow is a necessary condition for the single-eye special coordinate calibration method. We use the information of luminance, color, moving speciality to segment the target from background, and use some filter to remove the noise. The position of target is calculated with barycentre-method, outline-method and projection-method. We make use of the special geometry relation of the light, the ball, the shadow, and the lens-centre of the camera to obtain the 3-D coordinate of the ball. We build up the state function model of the flying phase of ping-pong and the motion model of the bounce process of ball, based on which we can predict the motion trace of the ball with a serial image. Getting the motion trace, we can move the robot to hit the ball. There are two key points in the project, time and precision. Based on the experiments, Grabbing images need 50ms, recognition and getting position need 100ms, coordinates calibration and motion trace prediction need 5ms. In all, vision takes up 160ms, during which the distance of the ball would have moved is not larger than lm. We can have enough time to move the robot and hit the ball. On the other side, the time of grabbing error is less than 1ms, the calibration error is less than 1cm, the motion trace error is less than 10cm, and the hitting time error is less than 30ms. With a view to the size of the racket and the range of the hitting movement, the ball can be hit back successfully.