Improved projection based M-estimators for fundamental matrix in stereovision
立体视觉中基本矩阵的改进投影M估计方法

In stereovision and image motion analysis, it is important to eliminate the mismatched feature points to get accurate fundamental matrix estimation. In order to overcome the limitations of kernel estimation and shortcomings of the method for determining inliers/outliers threshold in Error-In-Variables (EIV) model, the projection based M-estimators (pbM) was improved. First, a new mode search method named adaptive kernel density estimation was proposed. Then, a different strategy was given to determine the inliers. Finally, both simulation and experiments on real data validate the effectiveness of the improved pbM algorithm.