Outliers Elimination Based Ransac for Fundamental Matrix Estimation

Abstract

To accelerate the RANSAC process for fundamental matrix estimation, two special modifications about RANSAC are proposed. Firstly, in the verification stage, not the correspondences are used to verify the hypothesis but the singular values of estimated fundamental matrix are directly used to evaluate the effectiveness of the matrix. Secondly, after getting a plausible estimation, the obvious outliers are eliminated from the correspondences set. This process can enhance the inliers' ratio in the remaining correspondences set, which will accelerate the sample progress. We call our method as outlier elimination based RANSAC (OE-RANSAC). Experimental results both from synthetic and real data have testified the efficiency of OE-RANSAC.