利用单应变换与对极约束的基础矩阵估计算法  

Fundamental Matrix Estimation Using Homography Transformation and Epipolar Geometry

作　　者：佟强 王紫瑶 杨大利[1] 刘秀磊[1] TONG Qiang;WANG Ziyao;YANG Dali;LIU Xiulei(School of Computer, Beijing Information Science & Technology University, Beijing 100101, China)

机构地区：[1]北京信息科技大学计算机学院,北京100101

出　　处：《郑州大学学报（理学版）》2021年第1期61-67,共7页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家重点研发计划项目(2017YFB1400402,2018YFB1701602);国家自然科学基金项目(61771022);北京信息科技大学学校科研基金项目(1925019)。

摘　　要：为了提高不同图像之间的基础矩阵估计的精度与效率,提出了一种结合单应变换和对极几何约束的基础矩阵估计算法。在不同图像上提取特征点并建立匹配关系,过滤其中明显的误匹配点,并计算单应变换矩阵,由于单应变换的内点和外点在图像之间的对应关系均符合对极几何约束,通过将单应变换矩阵与外点结合,进一步计算出基础矩阵。为了验证算法的有效性,在采集的图像数据与公开的图像数据集上分别进行实验,实验结果表明,该算法与RANSAC相比,在内点率与内点平均误差上均有一定提高,具有更好的鲁棒性和有效性。In order to improve the accuracy and efficiency of fundamental matrix estimation between various images,a fundamental matrix estimation method was proposed by using homography transformation and epipolar geometry.This method firstly extracted several feature points and found their matching points in pairimage.The mis-matching point pairs would be discarded during this process.A homography matrix was found by the best point pairs,thus the points which were correspondence with these point pairs could be seemed as coplanar.Since all points and their projection pixels on different image planes had to follow the epipolar geometry,the fundamental matrix was estimated by the homography matrix and its outliers.The experiments on public and private image set showed that this method could achieve a more robust and accurate estimation of fundamental matrix.

关 键 词：多相机 基础矩阵 对极几何 共面性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]