A Fundamental Matrix Estimation Algorithm Based on Point Weighting Strategy

Estimating Fundamental matrix from corresponding points is an important problem in the field of computer vision. The random sample consensus (RANSAC) is one of the most effective methods for Fundamental matrix estimation. In this paper a point weighting strategy is added to RANSAC in order to improve the efficiency. The algorithm gives initial weight to each point, and the weight of corresponding points are changed according to the evaluation value of fundamental matrix computed in each sampling. The weight of points will affect the probability of points to be extracted, and the inliers which have a larger weight than outlier will be more likely to be extracted. The fundamental matrix computed in each sampling is evaluated by the weight of corresponding points, and the weight of the corresponding points are updated in turn, so the whole process forms a positive feedback. Experimental results on synthetic data and real images demonstrated that the new algorithm is valid and robust.