Precise fundamental matrix estimation based on inlier distribution constraint

The fundamental matrix is an effective tool to analyze epipolar geometry relationship between two-view images and plays an important role in computer vision. Traditional RANSAC method selects the biggest consensus set of inliers to estimate fundamental matrix. No previous methods have considered whether such a choice really is the best. In this paper, a new algorithm for fundamental matrix estimation by considering the inliers distribution is proposed. It takes the traditional RANSAC method as the basic framework and selects these sets which contain a large number of inliers to construct a candidate set. Then calculate the density of the inlier distribution and the mean of the epipolar distance of the inlier sets in the candidate set. At last choose the optimum one as the inlier set to estimate the fundamental matrix. Through experiment comparison with previous methods on a large number of simulated and real image data show that the proposed algorithm can achieve a more precise result. © 2013 Springer-Verlag.