Fast approach of randomized detection for multi-ellipse

Abstract

For multi-ellipse detection, this paper proposed a randomized fast detection approach. It made use of randomly sampled one edge-point sampling randomly and locally searched two edge-points in an image and information of their neighborhoods to determine a candidate ellipse, then affirmed true ellipse for the candidate ellipse by affirming true circle for corresponding circle into which is transformed by the candidate ellipse. When determining a candidate ellipse, the number of points sampled randomly was maximally reduced, more non-ellipse points were eliminated, and therefore useless sample and calculation were reduced. Numerical experiments demonstrate that this approach has better robustness and can advance the detection speed compared with the existing algorithms.