机器视觉中矩形几何提取和校正

作者：王红军，陈临强*，王磊 日期：2010-06-28/span> 浏览：3384 查看PDF文档

机器视觉中矩形几何提取和校正
王红军，陈临强*，王磊
（杭州电子科技大学 图形图象研究所，浙江 杭州 310018）

摘要：为了提高从图像中提取矩形的效率和精度，提出了一种实用的基于Hough变换和最小二乘拟合相结合的矩形几何提取和校正算法。该算法通过Hough空间累加器的最大值检测出直线；在检测到的直线邻域内，通过最小二乘法拟合得到更精确的直线；由4条直线得到4个交点，并按各个点与中心点的位置关系进行排序；由4对对应顶点，可求出一个透视变换矩阵；最终将四边形内的所有点映射到一个矩形区域内。实验结果证明算法在效率和精度上取得了一个较好的平衡。
关键词：Hough变换；最小二乘法；矩形检测
中图分类号：TP391文献标识码：A文章编号：1001-4551(2010)04-0095-03

Extraction and correction of rectangle geometric date in machine vision

WANG Hongjun, CHEN Linqiang, WANG Lei
（Institute of Graphics and Images Process, Hangzhou Dianzi University, Hangzhou 310018, China）

Abstract: To improving the efficiency and accuracy in rectangle extraction from the image, a practical algorithm about rectangular geometry extraction and calibration was presented based on Hough transform and leastsquare algorithm. A straightline was detected, using the maximum data stored in Hough space accumulator. Then a more accurate line was fitted in the detected line region by using leastsquare algorithm. Four intersection points gotten from four lines were sorted, according to their position`s relationship from each point to the central point. A perspective transform matrix was computed through four pairs of points. All those points in a quadrangle were mapped to a rectangle by multiplied the matrix .The experiment results prove that, this method can achieve a good balance in efficiency and accuracy.
Key words: Hough transform; leastsquare algorithm; rectangle detection