Outliers handling in the structure-from-motion algorithm

Structure from motion (SfM) is an important task in computer vision. Nowadays, most SfM algorithms are based on feature points, while the outliers from the feature points extraction will limit the accuracy of the SfM algorithms. To eliminate outliers, the RANSAC-style methods are widely used but they have high time complexity. In this paper, a soft-weight-assign method is proposed for the iterative SfM methods to handle outliers. Simulation and experimental results show that the effect of outliers can be strongly reduced by the proposed soft-weight-assign method.