A fast contour-based approach to circle and ellipse detection

The real-time detection of circle and ellipse from complex background is a very challenging problem. This paper presents an efficient and effective method introducing the connectivity constraint which dramatically enhances the performance of detection and accuracy of localization. Lacking in abstract denotation of features, most of the current techniques suffer from the huge feature point set. In contrast to presenting features as isolated points, we propose to regard features as contours, continuous lines of one-pixel width. The chain code algorithm is employed to organize feature points as disjoint feature contours. For each independent contour, the parameters of possible figures (circle/ellipse) are estimated based on the RANSAC (Random Sample Consensus) algorithm, thus reducing the scale of problem to several smaller subproblems. To avoid the influence of small arcs and to avoid the case that several targets overlap in the real application, the 'fitting factor' of each possible figure is introduced to evaluate the estimation of parameters. Our experiments successfully demonstrate its real-time performance, accuracy, and robustness.