基于自适应Ransac算法的基础矩阵估算

Estimating of Fundamental Matrix Based on Adaptive Ransac Algorithm

基础矩阵估算是计算机视觉中恢复两张影像相对运动不可缺少的步骤,通常使用Ransac算法从包含大量外点的数据集中估计基础矩阵。经典Ransac判断样本是否为内点的阈值是经验值,不能满足所有样本,且在两个矩阵含有相同内点数时无法确定最优基础矩阵。本文介绍了自适应Ransac算法及流程,将其应用与基础矩阵的解算中。通过利用无人机影像与手机拍摄照片进行实验。实验结果表明,自适应Ransac算法可得到比经典Ransac更精确的基础矩阵。

Estimating the fundamental matrix is an indispensable step to recover the relative motion of two images incomputer vision technology. The Ransac algorithm is usually used to estimate the fundamental matrix from the data set containing a large number of external points. The threshold of Ca^sical Ransac that determines whether a point is in-ternal point is the empirical value,whichcan？t satisfy all the samples, and the optimal fundamental matrix can？t be de-termined when the two matrices contain the same number of interior points. This paper introduces the Adaptive Ransac algorithm and flow path, apply t to the solution of fundamental matrix. Experiments are carried out by using UAV images and mobile photographs. It is proved that the Adaptive Ransac algorithm can inl^erred more accurate l^un- damental matrix than the Classic Ransac.