摘要
通过分析基本矩阵的鲁棒估计方法的特点,提出了三点改进:在RANSAC(RANdomSAmplingConsensus)方法中采用了极小化再投影误差判别数据点的类别;给出再投影误差的一阶近似算法;由求出的基本矩阵和局内点数据采用LM算法对结果过一步求精,给出更好的基本矩阵估计值,使得再投影误差进一步减小,避免结果趋于局部极值。合成数据和真实图像实验均证明了该方法的有效性和可靠性。
After analyzing the characteristics of methods for computing fundamental matrix, a method was presented for robustly estimating fundamental matrix with three improvement. The data set was discriminated into iuliers or outliers by minimizing reprojection error. The computation of one order approximation for reprojection error was given. To avoid local minimum, LM algorithm was adopted in last steps of RANSAC（ RANdom SAmpling Consensus） algorithm. A very good estimation of fundamental matrix was obtained and the reprojection error was smaller. Experiment results on synthetic and real images demonstrated that the new algorithm is valid and robust.
出处
《计算机应用》
CSCD
北大核心
2005年第12期2845-2848,共4页
journal of Computer Applications
关键词
基本矩阵
再投影误差
LM算法
RANSAC方法
fundamental matrix
reprojection error
LM algorithm
RANSAC（ RANdom SAmpling Consensus） algorithm
作者简介
郭继东（1973-）,男,山东济南人,讲师,硕士,主要研究方向：三维重建技术、计算机视觉、计算机图形学;（Gjd730210@163.com）.
向辉（1970-）,男,山东济南人,副教授,博士,主要研究方向：图形图象处理、三维重建.
