Multi-level cross-view line matching: exploiting spatial domain characteristics

Cross-view line matching is critical for high-resolution urban modeling, but existing methods rely on priors and struggle with curves. Thus, we propose a multi-level cross-view line matching method by spatial domain characteristics. It employs fast affine simulation to compensate for distortion and introduces the perspective invariance of coplanar line intersections to obtain multi-type corresponding line pairs and line features. Then, local homography constraints are established to achieve single-line grouping and matching. Experiments demonstrate that our method outperforms baselines in matching accuracy and efficiency, and supports curve matching, providing an efficient approach for reconstruction based on irregular contours.