Abstract: It proposes a new method to detect automatically badminton appearance. First, images of badmintons were collected and processed in a series of steps. Then, by using the method of region-dependent segmentation, based on polar-coordinate, the feature points of binary images were located. Thirdly, the least-square method was adopted for fitting. Finally, the related indices were propounded to assess roundness errors. The experimental results prove that this method is feasible to differentiate badmintons of the C and D grades, making the nontouched on-line measurement possible.