A real-time and precise ellipse detector via edge screening and aggregation

Abstract

A fast and precise method for ellipse detection is proposed in this paper. The method aims at clearly removing the lines and curves which are not ellipse edges to improve the ellipse fitting. In arc extraction, the arcs are divided into four categories according to the gradient, and the size constraint is exploited to remove the interference lines. Then, the arc relative position constraints and the tangent lines constraint are employed to exactly group the arcs that belong to the same ellipse into a set. Finally, a post-processing approach is developed to remove the invalid ellipses. Due to the effective removal of the interference edges and the designed geometric multi-constraint, the computational costs of arc grouping and parameter estimation are dramatically reduced, and the fitting results are finely agreeable to the actual ellipse contours. The performance is evaluated with 3600 synthetic images and 1517 real images, and the experimental results demonstrate that the proposed method runs much faster than the current speed leading methods with the comparable or higher F-measure.