Abstract: This method searches respectively an edge point on three line segments regarding two edge points sampling after two edge points are picked randomly. A possible ellipse is determined with two sampling points and three searching points. When sampling two edge points and searching three edge points, isolated noises, half-link noises and edge points not being on the same true ellipse are recognized and eliminated, as decrease significantly invalid sample and invalid computation. After founding possible ellipse, this method transforms the possible ellipse into corresponding circle and applies an evidence-collecting process to determine whether possible ellipse is true and to control directly detection accuracy of the ellipse. The experimental results demonstrate that the method can perform better than other algorithms of multi- ellipses detection on speed and robustness.

Key words: Generalized Hough Transform, Multi-ellipse Detection, Line fitting

摘要： 本文随机采样2个图像点后再分别在这2点相关的3条线段上搜索1个图像点。利用采样的2个图像点和搜索的3个图像点决定侯选椭圆。当采样和搜索图像点时,通过识别和剔除孤立、半连续噪声和不在同一个真椭圆上的图像点,显著地减少了无效采样及无效计算。在确认真椭圆时,该方法将椭圆变换成对应圆,通过确认真圆来确认真椭圆并直接控制椭圆的检测精度。数值实验结果表明:本算法在检测多个椭圆时的检测效率和鲁棒性等方面都具有良好的性能。

关键词: 广义Hough变换, 多椭圆检测, 直线拟合