keyword

correspondence search

Correspondence search is the process in computer vision and three-dimensional reconstruction of identifying and matching points across multiple overlapping images that represent the same physical location in a scene. This procedure serves as the primary stage in structure-from-motion workflows and typically encompasses feature extraction, where distinctive keypoints and visual descriptors are detected; feature matching, where similar descriptors between image pairs are paired; and geometric verification, which eliminates incorrect matches using epipolar geometry and robust estimation models. The resulting set of geometrically verified correspondences and connected image tracks forms the foundational data structure used in subsequent stages to recover camera parameters and reconstruct three-dimensional spatial coordinates.

1 item