earticle

영어

To solve the problem that the conventional circle detection algorithm has low processing speed and poor identification accuracy in presence of overlap of objects, a novel multiple circles detection algorithm is proposed. Firstly, the consecutive edges with single pixel are obtained through an edge tracking method. Moreover, the edges of overlaid objects are extracted from the cross section according to the linear projection method. Secondly, a large number of non-circular edges are excluded by utilizing a coarse detection that decides whether four points on an edge are on the same circle or not. Lastly, the circles are recognized by the fine detection according to the geometrical feature which two subarcs belonging to one circle have the same center and radius. The experimental results demonstrate that the algorithm is superior in speed, precision and accuracy, comparing to the conventional ones, such as Hough transform circle detection and randomized circle detection. Furthermore, it can solve the issues from overlap of objects and accurately identify the multiple circular objects in the complex scenes as well. The algorithm is concise, fast and robust.