A new mismatching elimination algorithm is proposed for the ORB algorithm

aiming at the problem of the traditional RANSAC algorithm in eliminating the mismatch.The algorithm divides the elimination stage into two parts: rough elimination and fine elimination.The rough elimination selects the appropriate Hamming threshold size to eliminate some mismatching pairs by experimental design

so as to reduce the calculation amount of fine elimination stage.The fine elimination uses the information of the feature point and its neighbourhood pixelpoints to construct the difference matrix

and defines the singular value of the matrix as the description vector of the feature point.The effect of further elimination is achieved by calculating the correlation of the two vectors.Experimental result shows that the algorithm can effectively eliminate the mismatching point pair

and the precision is improved by 8 % and the matching precision is 98.97 % compared with the RANSAC algorithm.