A Compound and Robust Algorithm for Ellipse Detection

Abstract

Aiming for ellipse detection in complex environment, we propose a compound algorithm. In the scene image that we see everyday, there are usually many corner points and straight lines and it is not practical to use Randomized Hough Transform (RHT) to detect ellipse from such an image, for that the corner points and straight lines everywhere bring numerous noneffective samplings and accumulatings. Aiming at the solution of the problem, we firstly filter noisy points, corner points and straight lines, as many noneffective samplings are eliminated, then we use a compound ellipse detection algorithm to detect ellipse. Firstly use all points of the curve to fit ellipse by least squares and judge if it is the right ellipse, if not, sample five points random from the curve to solve the ellipse parameters, then an effective ellipse fitting rule is proposed to judge whether a point belongs to the solved ellipse. We use the above random sampling and ellipse fitting rule repetitiously to find the most fitting ellipse. In above processing we make full use of the continuity of the edge to sample points random and fit ellipse, as it reduces much more noneffective samplings and accumulatings. Simulation and experiments indicate that this algorithm is more robust and faster than RHT.