%0 Journal Article
%T 基于梯度分类的复杂背景椭圆快速检测方法
%A 吴晨睿
%A 张树有
%A 何再兴
%J 浙江大学学报(工学版)
%D 2018
%R 10.3785/j.issn.1008-973X.2018.05.014
%X 针对复杂背景下椭圆特征由于重叠、缺失、嵌套等原因导致的检测效率低、误检率高的问题，提出基于梯度分类与多边形辨识的椭圆快速检测方法. 该方法通过边缘检测算子对采集的图像进行预处理，获取图像边缘的梯度信息. 根据边缘灰度梯度与凹凸性将边缘线分为4类圆弧特征，通过对4类圆弧特征的聚类初步确定备选的椭圆集合. 利用椭圆内包多边形为凸多边形的特点，对候选椭圆集合进行快速辨识. 应用非迭代几何最小二乘法拟合椭圆参数，通过椭圆残差判定与椭圆的去伪过程，获得最终的椭圆特征. 实验结果表明，该方法在椭圆检测效率与准确性上较经典算法均有提升.</br>Abstract: A novel ellipse detection method based on gradient clustering and convex polygon was proposed in order to solve the problem of slow detecting speed, low accuracy and high error rates of ellipse detection in complex background. Edge information including location and orientation was extracted through Canny operator in image preprocessing. Edges were clustered into four categories according to their gradient orientation and convexity. Unqualified ellipse tribes were quickly filtered according to convex polygon property. A non-iterative geometric least square method was used to fit the ellipse. Ellipses with small fitting errors were confirmed to be the final ellipse features in complex background. Experimental results show that the proposed method performs better than the classical algorithm in both recognition accuracy and calculating time.
%U http://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2018.05.014