On-board Vision System For Humanoid Ping-pong Robot Without Marks And Trajectory Prediction

Vision system has always been one of the difficult problems of ping-pong robot research. The vision system for the existing humanoid ping-pong robots is mostly using the solution of external fixed vision system, not the solution of on-board vision system. This thesis is based on humanoid robots of Zhejiang University and research how to make humanoid robots based on on-board vision system to play ping-pang with human beings without additional marks on the table. In addition, in order to realize trajectory prediction of the ping-pong ball in on-board vision system, this thesis analyzes the parameter learning between external fixed vision system and on-board vision system. The main contributions are listed as follows:1. Based on self-localization using points, this thesis discusses how to use the information on the table to realize self-localization and proposes a line-recognition algorithm based on prediction to track line in real-time when the on-board vision system is dynamic. The algorithm accuracy is about 99%,so the vision system can detect lines well even the on-board vision system is dynamic.2. This thesis proposes a pose estimation algorithm from door-like lines in a two-camera system. Campared with the pose estimation algorithm in single-camera system, the pose estimation algorithm from door-like lines in a two-camera system has the advantage in stability and precision.3. This thesis proposes an algorithm to transfer the parameter between external fixed vision system and on-board vision system. When the speed of ping-pong is too fast, the visual system don’t have enough time to predict trajectory, this thesis proposes an algorithm based on collision model to predict trajectory in advance. In addition, this thesis discusses the relationship between the baseline and accuracy of positioning. Humanoid robots based on on-board vision system have made 28 rounds with human beings.