Fast Randomized Algorithm for Circle Detection by Efficient Sampling

According to the randomized circle detection (RCD) algorithm, the probability that four randomly sampled points belong to the same circle is relatively low. A fast algorithm for circle detection in this paper is proposed. Three points will be randomly sampled in the edge image and the fourth point is then sampled within the margin of vertical circumscribed and inscribed squares of the circle determined by three previously sampled points, and it effectively increases the probability of four randomly sampled points belonging to the same circle. In judging whether these four points can determine a candidate circle, a threshold is defined to reduce the calculation times of circle parameters. It affirms the candidate circle for true circle by a fast evidence-collecting process. Experimental results demonstrate that the proposed algorithm is about an order of magnitude faster than RCD in the detection speed, and has a strong robustness. The detection precision for circles is better than half a pixel.