An efficient circle detector not relying on edge detection

Accurate and efficient detection of circular modules fixed on non-cooperative target is a key technology for Tethered Space Robot. This paper presents an efficient circle detector based on region-growing of gradient and histogram distribution of Euclidean distance. Region-growing of gradient is applied to generate arc support regions from single point. And the corresponding square fitting areas are defined to accelerate the detection and decrease storage. A histogram is then used to count frequency of the distances that participates in the accumulator and the parameters of each circle are acquired. Finally, a verification strategy of circular integrity is designed to test the detection results. We have tested our algorithm on 35 images dealing with kinds of circles and ellipses. Experimental results demonstrate that our method is able to detect circular objects under occlusion, image noises and moderate shape deformations with a good precision.