Error matching elimination based on a local affine algorithm

Abstract

Abstract Aiming at the shortcomings of orb algorithm using the traditional RANSAC (random sampling consistency) algorithm to eliminate mismatches, such as the high number of iterations when the sample size is large, and the proportion of error points also affects the fitting accuracy of the traditional algorithm, a rough and fine method to eliminate mismatches is proposed; The threshold screening method is used to select the optimal threshold and preliminarily eliminate a large number of wrong matching pairs to achieve the effect of rough elimination; Then the sub-window method is used to calculate the local affine matching under all windows, reduce the number of error points, and increase the correctness of the fitting model of RANSAC algorithm. The experimental results are compared with the traditional algorithm, and the matching accuracy is improved by 6% and combined with orb-slam2.