In order to obtain robust fundamental matrix from video images

a robust estimation algorithm for fundamental matrix suitable for video frames is proposed.By using three consecutive frames as a group and pairwise matching between successive frames respectively

the optimal and better two sub-sample sets are determined.Random sampling is conducted in the better sub-sample set

and the initial fundamental matrix is obtained by using the new preliminary test method.Combining the correspondence score and the inlier rate as the weight factor

the fundamental matrix is estimated by weighting.The two epipolar lines corresponding to the intermediate frame of the first frame and the third frame intersect at the same point to form a constraint relationship

and the two fundamental matrices to be optimized simultaneously by bundle adjustment.Experiments show that this method is more accurate and robust than traditional methods.