基础矩阵估计的加权归一化线性算法

基础矩阵是对来自同一景物的两幅未定标图像进行分析的基本工具.对其进行估计的常用线性算法有八点算法和改进的八点算法,其最大的优点是运算简单、易于实现,但对噪声和错误数据较敏感,因此实用性差.通过引入与余差有关的代价函数,给出了一种新的鲁棒性线性算法——加权归一化算法.首先将原始输入数据加权归一化处理,然后再用八点算法求F阵的8个参数,实现了F阵的估计.大量的模拟数据和真实图像的实验结果表明,此算法不仅具有良好的鲁棒性,而且可提高基础矩阵的估计精度.

A Weighted Normalization Algorithm for Estimation of Fundamental Matrix

The fundamental matrix is a basic tool in the analysis of scenestaken with two uncalibrated cameras. The 8-point algorithm and the improved 8-point algorithm are widely used linear methods for estimating the fundamental matrix. They have advantages of simplicity in implementation. But they are extremely sensitive to noise and outliers. Hence in most cases, they are useless virtually. A new robust linear method——weighted normalization algorithm is developed by introducing a cost function related to residual errors. Firstly, the matching points with a weight factor are normalized. Secondly, the eight parameters of fundamental matrix are calculated by using the 8-point algorithm. Experiments on simulated and real image data are conducted. The results show that this algorithm is very robust to noises and outliers, and the fundamental matrix with high accuracy can be found.