Transport by robotic throwing and catching: Accurate stereo tracking of the spherical object

Robotic catching give the specific requirements to the observation system, which is used for tracking the thrown object. Here the development of the image processing algorithm for real-time positioning of the thrown tennis ball is described. Influence of the camera positioning and chosen image processing algorithms (background subtraction, Hough and RANSAC circle detection) on positioning accuracy is discussed. Accuracy of positioning is analyzed for two modes of the vision system: learning mode and working mode. In the learning mode the there is no hard time constraint and entire trajectory must be accurately reconstructed. In working mode the observation must be real time and only first three quarters of the trajectory is observed. The algorithms of accurate processing for both modes is developed. The efficiency of the observer is proved by the experiments in robotic catching.