Study On Real-time Vision Based Calibration And Trajectory Tracking Technology Of Ping Pong Robot

Due to the high intelligence and broad application fields, robotics technology is an important indicator of national high-tech and industrial automatic level. In recent years, a novel ping pong robot has gradually become one of the research hotspot by reason of its difficulty and challenge in both real-time and intelligent design. As one part of the "high performance humanoid robot" project which is supported by National High-tech R&D Program (863 Program), our lab has developed three generations of ping pong robot till now, and we aspire to catch up with the research process of United State, Japan, Europe and other developed countries.The development of vision system which is equivalent to the "eyes" of the ping pong robot is a top priority task among all the key techniques. This dissertation, based on the groundwork of these three generations of ping pong robot, mainly focuses on technical difficulties during designing different functional modules of vision system. Real-time vision based technical points involved in the process of object recognition and high-speed moving targets tracking, such as synchronous acquisition of the image data, monocular and binocular camera calibration, ping pong ball trajectory tracking and prediction and real-time motion planning, were explored and innovated in details.Both solutions on multiple camera synchronous acquisition and compensation of long exposure time were well established. Fast filtering and object recognition method were used to resolve the "motion blur" phenomenon caused by high-speed moving objects'imaging. ROI dynamic prediction algorithm was introduced to satisfy the accurate real-time requirement in the image processing and target tracking. In the monocular vision model constructed with "light", "shadow", "camera" and "ping pong ball", according to the solid geometric constraint, principle of pinhole and perspective projection, an optimized lease square calibration method was applied, which not only calibrated the parameters of the camera, but also the parameters of light. Further more, the method of locating the ping pong ball in three-dimentional space under stereo vision and camera calibration method of binocular vision were studied. Kinematic model for the ping pong trajectory prediction and physical model for the ping pong collision were further analyzed. Firstly, in order to resolve the measurement error and modeling error occurred in the process of ball trajectory tracking, an adapted variable covariance fast tracking method was proposed; this method turned out to guarantee the real-time requirement, convergence and stability of the tracking performance. Furthermore, taking into account the actual existence of non-linear air drag term in the ping pong movement, an online "drag coefficient" extended Kalman filter tracking method was developed based on the physical modeling of ping pong ball; this method was confirmed by its higher speed and accuracy to achieve a faster and better tracking result of ping-pong ball. Besides, an aerodynamics model of ping pong ball, a simulation on Magnus force and spinning ball trajectory detection were initially explored.In this dissertation, real time vision technology advanced its application as well as innovation on the ping pong robot, which is also a valuable reference for relative research such as high speed visual servoing and fast target tracking.