On the Automatic Estimation of Fundamental Matrix

Accurate and robust estimation of the Fundamental matrix is very important for many computer vision applications, such as 3D reconstruction from 2D images. In this paper, a complete analysis and comparison of three important algorithms for estimating the Fundamental matrix when three views are given. The six-point algorithm originally used for three-view geometry is implemented for the automatic estimation of the Fundamental Matrix — the two-view geometry tensor, assuming that more than two views are available. An extensive comparison to other two popular methods: the seven-point and eightpoint algorithms are made through carefully designed experiments with synthetic and real images, assuming three views are available. The comparisons show that the new-implemented six-point algorithm, which makes use of the three-view constraint, is more accurate and has higher ability to handle the outliers.