Detection Of Partially Occluded Ellipses

Elliptical curves or fragments of such curves emerged in many industrial applications, Ellipsedetection is therefore a powerful method of cuing into these specific geometric object models.Traditional ellipse detection algorithms based on Hough transform is not only have a burdenof computation but also lack robustness under occlusion. Both of the problems greatly limitthe application of ellipse detection.One shortcoming of traditional algorithms is that they have not considered the globalproperties of ellipse edges. The information they used were constrained in the points-pairs,triplets or the tangential and gradient of edge points. Our motivation is to incorporate moreglobal properties into the ellipse detection with the newly-developed conic fitting methods.We find the common characteristic of elliptical segments and try to use less edge informationto recover the whole ellipse.Our works mainly lie in the following aspects: (1) Divide the ellipse detection algo-rithms into two classes and point out no algorithm alone can be robustly implemented undermoderate occlusion; (2)Integrate the conic fitting methods into Hough transform and pro-pose a new algorithm with advantages from both sides. It not only reduces the computationburden but also solve the information lost problem due to occlusion. Experimental resultsdemonstrate that our method can retain a reasonably high accuracy under moderate occlu-sion (such as 1/3). Moreover, it also shows robustness to the orientation of ellipses; (3)Wegives out a quantitative and qualitative analysis about the main parameters on which the per-formance depends and point that all the parameters'effects are ascribed by the noise effectwhich disturb the location of edge points accurately.