摘要
针对特征定位误差服从各向异性高斯分布时单应矩阵的优化问题,提出了一种归一化椭圆权重的烈文博戈马奎特(EW L-M)算法。该算法对目标函数添加椭圆权重的形式,使得目标函数在马氏距离的基础上兼有欧式距离优点。模拟与真实数据的实验结果表明,本文提出的方法能在更少的迭代次数下估计出更精确的单应矩阵,且对不同级别的误差表现出更强的鲁棒性。
Homography estimation is a hot topic in the field of computer vision. In this paper, a normalized ellipti- cal weight Levenberg-Marqardt (EW L-M) algorithm is proposed as a way to solve the homography optimal prob-lem when the feature's location error has an anisotropic Gaussian distribution. By adding elliptical weights to the ob- jective function, the algorithm can avoid losing the advantages of Euclidean distances when using Mahalanobis dis- tances. Experiment results with simulated and real images show that the EW L-M algorithm can estimate a homog- raphy matrix with greater accuracy in fewer iterations and displayed strong robustness for different levels of error.
作者
毛路路
祝海江
MAO LuLu;ZHU HaiJiang(College of Information Science & Technology, Beijing University of Chemical Technology, Beijing 100029, China)
出处
《北京化工大学学报(自然科学版)》
CAS
CSCD
北大核心
2019年第3期83-86,共4页
Journal of Beijing University of Chemical Technology(Natural Science Edition)
基金
国家自然科学基金(61672084)