广义Hough变换在遮挡图像识别中的应用

Applications of the generalized Hough transform in recognizing occluded image

介绍了一种基于广义Hough变换的空间有部分遮挡的茄子目标识别方法。从生物学的角度,用广义柱近似等效样本茄子形状;通过坐标转换获得了不同位姿的立体目标在平面上的投影边缘,并建立茄子外形的4维参数索引表;采用一种广义Hough变换计算茄子目标的潜在位置,并通过比较各旋转角度下的"面积比",筛选出目标实际的位姿。试验表明:这种广义Hough变换检测目标位置和姿态的相对误差分别约在1.3%和5.5%以下,说明其对空间不同位姿、部分遮挡情况下目标的识别具有良好的效果。

This paper introduced a method of recognizing the partially occluded eggplant based on generalized Hough transform in three-dimensional space.The shape of eggplant was approximately described by using generalized-cylinders from the view of biology.Different edges of projection were obtained through coordinate transformation and 4-D parameter index tables were established.Generalized Hough Transform was adopted to count the potential pose of object and the real pose was screened by comparing the area-ratio of different rotation angles.The experiment demonstrates that the relative errors of position and pose detecting for target are lower than 1.3%and 5.5% by using Generalized Hough Transform,and this method is feasible and effective to recognize different poses and partially occluded objects.