Abstract: In view of the high redundancy and poor globality of the same-name point pairs when the BRISK feature detection algorithm matches in remote sensing images, this paper considers that the BRISK feature detection algorithm can obtain a large number of UAV image feature points, and the Delaunay triangulation algorithm can use the rough matching point pairs of the BRISK feature points of the image construct a triangulation network. Combining the advantages of the two algorithms, a method combining the BRISK feature detection algorithm and the Delaunay triangulation algorithm to eliminate mismatched point pairs of UAV images is proposed. This method uses the BRISK rough matching feature points of the two images to construct the Delaunay triangulation, uses the triangle similarity in the traversal of the two images to eliminate the mismatching point pairs, and then uses the photographic invariant principle to further eliminate the wrong matching point pairs, improving the accuracy of the image matching. This paper compares and studies the effect of the projective invariant algorithm of Delaunay triangulation and the RANSAC algorithm to eliminate the original image group, adds the pepper-salt noise image group and the rotated image group to the effect of the BRISK feature mismatch point pairs. The experimental results show that the three sets of images respectively use the UAV remote sensing image matching method combining the BRISK feature and the Delaunay triangulation's projective invariant algorithm to obtain the correct feature matching points with low redundancy and excellent global performance.

Key words: UAV remote sensing images, BRISK feature, RANSAC algorithm, Delaunay tri-angulation, projective invariants, triangle similarity