中文摘要：

以随机采样一个图像点P1的5×5邻域图像点作拟合直线l1，与l1，距离为d（d〉0）的平行线l3与l2（l2是通过P1点并垂直于l1的直线）的交叉点为Q，然后以Q为起点，在直线l3上按给定规则搜索两图像点P2和P3，用P1、P2和P3来确定候选圆。当采样和搜索图像点P2、P3时，通过剔除孤立、半连续噪声和非共圆点显著地减少了无效采样和无效计算。数值实验结果表明，该算法能快速检测多个圆。在检测多个圆时，其检测速度比随机圆检测算法快一个数量级；在孤立和半连续噪声不低于所有噪声的80％时，其检测速度比多个圆的快速随机检测算法大约快20％。

英文摘要：

An edge point P1 was picked randomly after isolated noises and half-link noises were eliminated. A fitting straight line l1 was made with all edge points of 5 × 5 neighborhood of the P1. A straight-line l3 that parallels to line l1 and which was d（d 〉0） apart from line l1 and intersects line l2 at Q was created（where l2 was line perpendicular to line l1 and through P1 ）. Beginning from Q, toward two ends search edge point P2 and P3 on line l3 respectively, It determined one possible circle using P1 , P2 and P3, When searching point P2 and P3, the non-co-circle points were recognized by character of the circle, and affirm quickly the possible circle for true circle in minimal area so that invalid computation was decreased. The experimental results demonstrate that the approach can detect quickly multiple circles, The detection speed of multi-circle is faster an order of magnitude than randomized circle detection（RCD） and about 20% than fast randomized multi-circle detection when number of isolated and half-link noise points account for upwards of 80% of all noises.