The Ellipse Detection Based On Deep Network Framework

Recently,detecting ellipse from a picture fast and accurately has been a necessary research question and needed for many target detection researches.Unfortunately,many problems still exist in the ellipse detection,e.g.,the multiple ellipse problem,the approximate ellipse problem,the nesting ellipse problem,the incomplete ellipse problem,the small ellipse problem and so on.Limited by these problems,exploiting the related ellipse detection applications are difficult.This paper proposes a well-designed ellipse detection algorithm based on deep network framework.In detail,we utilize the target detection algorithm of the deep learning to box out the ellipse of a picture,through which we can remove noise and make the bounding box tangent to the ellipse with the most possible.Then our following works are carried out based on the calculated bounding box.Meanwhile,to overcome the above mentioned problems,we take some measurements.Firstly,to reduce the arc matching process and then significantly enhance the detection speed,we detect ellipse in each single bounding box of a picture.To solve the nesting ellipse problem,we limit the detection area of the arc detection.To solve the approximate ellipse problem where the traditional methods are ineffective,we firstly use the bounding box to find out the position and then relax the verification condition.To solve the incomplete small ellipse problem,according to that the bounding box is almost tangent to the ellipse,we draw an ellipse directly when the bounding box is less than a certain threshold.As a result,the algorithm can effectively solve some problems which the traditional methods cannot deal with.In this paper,a lot of experiments are carried out on the data sets of three realistic pictures,which proves that the algorithm can effectively solve the existing problems of the ellipse detection,and has achieved a similar effect to the traditional ellipse detection algorithm in terms of accuracy.