A Novel Method of Stereo Camera Calibration for Tracking Baseball Pitches

Abstract

In this thesis, a stereo camera system for 3D reconstruction of arbitrary baseball trajectory is developed and a practical and flexible stereo camera calibration method is proposed. Conventional stereo calibration methods use a calibration object or distinguishable markers placed in the view, which makes them impractical and inconvenient for large spaces and wide baseline cases. The feature based stereo calibration method works well, but it is unable to extract enough point correspondences in extreme cases such as wide baseline. A more practical method is proposed to calibrate the wide baseline stereo camera system that uses several arbitrary baseball trajectories, from different point of views, as the point correspondences. After obtaining sufficient good spatially distributed trajectories, the Gold Standard algorithm is used to estimate the fundamental matrix and the camera pose. As a result, the stable 3D metric reconstruction of baseball trajectory from pitcher place to home plate can be done with high precision.