The 16th Meeting on Image Recognition and Understanding High Accuracy Ellipse-Specific Fitting

Detecting circles and ellipses in images is the first step of many computer vision applications including industrial robotic operations and autonomous navigation. To this end, point sequences constituting elliptic arcs are detected by image processing operations, and then an ellipse equation is fitted to them. In the past, many methods have been proposed for ellipse fitting [1], [3], [4]. However, most of them fit a quadratic equation in x and y, or a conic, to a point sequence. Usually, this produces an ellipse if the sequence forms an elliptic arc, but a hyperbola or a parabola could result when the sequence is very short and the noise is very large. It is Fitzgibbon et al. [2] who first proposed a method that only fits an ellipse. It is an algebraic method, and the computation is very easy, but the accuracy is low. Recently, Szpak et al. [5] introduced a high accuracy ellipsespecific method based on Sampson error minimization. In this paper, we incorporate a procedure for avoiding nonellipses to the hyper-renormalization method of Kanatani et al. [3], which is regarded as the most accurate of all currently known methods and demonstrate by experiments that our technique outperforms the method of Szpak et al. [5].