In pattern recognize and computer vision, circle detection is a critical issue. For the moment, the major concern of most circle detection algorithm is robustness and computational efficiency. Randomzied approaches for circle detection have the advantages of less computational time and less memory requirements. However, randomzied approaches involve in examining a large number of candidate circles and may not suitable in real-time applications. In the paper, the symmetry property of the circle was adopted to select the promising candidates for further investigation. At the same time, not all edge pixels in the image, but the edge pixels between the inscribed square and exo-square were collected when counting the number of pixels on the candidate circle. The experimental results show that the proposed algorithm performs fast while maintaining the accuracy.