Number ball recognition at arbitrary pose using multiple view instances

In this paper, we propose a novel strategy based on spherical feature extraction and fast polar image correlation to recognize number ball which has arbitrary rotation pose just using single calibrated camera. The multiple view instances of a number ball are collected as a prior knowledge of a class and a simple distance measurement is applied to decide which class the candidate ball belongs to. The proposed method is invariant to the three-dimensional rotation and scaling of the number ball, and it´s robust to illumination and noise. The experiment shows that our method is efficient and stable and achieves high recognition rate.