Abstract: A K-RANSAC ellipse detection algorithm is developed in this paper. By using the generalized orthogonality principle, all the properties of a circle is successfully generalized to the case of an ellipse. Then, by means of the affine transformation from a pair of arbitrary seed points to a pair of horizontal-vertical seed points, the algorithm can be extended to any case of the seed point pair. With normalization of the digital edge set, the ill-posed problem is effectively regularized. The generalized eigen analysis is employed to directly and effectively estimate the ellipse parameter vector. The detectability and fitting accuracy are significantly improved. Theoretical analysis and experimental results show that the proposed ellipse detection algorithm has the advantages of high robustness, wide applicability, high accuracy and good efficiency.