Vision system has always been one of the challenged problems in ping-pong robot research. The vision system for the existing ping-pong robot, especially for the humanoid ping-pong robot, is mostly using the solution of external fixed vision system, instead of the solution of on-board vision system. With the nowadays development of software and hardware technology, the vision system fixed on the body of robot will lead to three following challenges:one is the problem of dynamic due to the vision system shaking with the robotic body; Second is calculation in real-time to obtain both positions of robot itself and ping-pong ball; Third is the asynchrony of the on-board vision system.This thesis uses two humanoid ping-pong robots as experimental platform, and has done some researches on on-board vision system. The main contributions are as follows:1. A recognition algorithm based on prediction to track marking-points in real-time is proposed toward the on-board vision system is dynamic.2. A strategy for the accurate localization of robot in real-time, based on fusing two algorithms is proposed toward the on-board vision system is dynamic and asynchronous.3. A method to estimate ball trajectory accurately in real-time with asynchronous on-board binocular vision system for humanoid ping-pong robot is proposed. |