基于统计聚类随机椭圆检测方法  

The Statistical Clustering Method of Ellipse Detection

作　　者：唐进[1] 马树元[1] 吴平东[1] 卜庆红[1] 雷运洪[1] 

机构地区：[1]北京理工大学机电一体化中心,北京100081

出　　处：《微计算机信息》2007年第31期237-238,250,共3页Control & Automation

基　　金：国家自然科学基金(60274035)

摘　　要：本文提出一种基于统计聚类的随机椭圆检测方法。利用椭圆的几何性质和形状大小限定采样区域避免无效采样,获得候选椭圆时,检验边缘点集中位于候选椭圆上的点,通过统计候选椭圆上的点与其外切矩形上的点的个数的比值来验证候选椭圆的真伪。该方法与传统的Hough变换的参数空间累积验证相比,其计算简单,速度快,不需要为可能产生的虚假椭圆分配内存空间,更无需进行参数单元集的搜索,并能够快速的去除已验证的真实椭圆上的边缘点集,进一步减少了无效采样的产生。This paper proposed a statistical clustering method of ellipse detection. This method avoided many invalid sampling data which restrict the sample points utilizing the geometric properties and shape size of ellipse. The authenticity of a potential ellipse can be confirmed by the statistical value, which is the ratio of the perimeter of tangent rectangle of the ellipse to the amount of points with less distance to the ellipse than the threshold. Compared with accumulated validation in the parameter space of RHT（Randomized Hough Transform）, this method can simplify calculate at least cost, and it eliminates the need of allotting memory for the virtual ellipse and searching in the parameter space, moreover, it can quickly wipe off edge points in the validated ellipse and reduce invalid samnling.

关 键 词：椭圆检测 随机hough变换(RHT) 概率统计聚类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]