Considering the dissatisfactory precision and stability of primary M-Estimators

which depends entirely on the original matrix obtained by the method of least squares

an improved M-Estimators algorithm for estimating the fundamental matrix was studied The new algorithm obtained a more precise original matrix by calculating the distances between the matching points and the corresponding epipolar lines Then the mismatch and outliers in the original matching points set were eliminated through the precise original matrix and the nonlinear optimization for the new matching points set was carried out with Torr-M-Estimators Finally the accurate matching points set and the epipolar geometry can be gained Through a mass of experiments on simulated data and real images in the case of mismatching and Gaussian noise

the comparing results between the algorithm and other robust methods indicate the algorithm not only improves the estimating precision but also shows the good robustness