Abstract: In the process of UAV image matching, rough error is inevitable. Therefore, it is very important to obtain feature points with high robustness for UAV image matching. The traditional method is using classical RANSAC algorithm for coarse error elimination, which is affected by the sampling times, error threshold, and remaining partial mismatched points. By using graph theory, the feature points extracted by SIFT algorithm are preprocessed, that is the feature points with lower energy are removed by constructing the energy function of feature points, which can improve the robustness of matching feature points and reduce the feature points coarse error. The paper proposes a new method, which combines the graph theory algorithm with the classical RANSAC algorithm to eliminate the rough error. The method is named GSIFT-RANSAC algorithm, which can improve the robustness of feature points and obtain the homography matrix with high accuracy. Using different data for verification, gross error elimination rate of the algorithm proposed in this paper is 5.31% and 14.29% higher than the algorithm using graph theory to remove feature points alone, which indicates that the effect of proposed method is better.

Key words: UAV image matching, feature points, RANSAC algorithm, graph theory, GSIFT-RANSAC algorithm