On Making SIFT Features Affine Covariant

Abstract An approach is proposed for recovering affine correspondences (ACs) from orientation- and scale-covariant, e.g., SIFT, features exploiting pre-estimated epipolar geometry. The method calculates the parameters consistent with geometry point coordinates scales rotations which feature detector obtains. closed-form solver returns a single solution extremely fast, i.e., 0.5 $$\upmu $$ μ seconds on average. Possible applications include estimating homography upgraded correspondence and, also, surface normal each found in pre-calibrated image pair (e.g., stereo rig). As second contribution, we propose minimal that estimates relative pose of vehicle-mounted camera SIFT corresponding obtained from, ACs. algorithms are tested both synthetic data number publicly available real-world datasets. Using solvers leads to significant speed-up homography, multi-homography estimation problems better or comparable accuracy state-of-the-art methods.