Research news on structure from motion

Structure from motion (SfM) is a computational method in computer vision for reconstructing 3D structure and camera motion from a set of overlapping 2D images acquired from different viewpoints. SfM jointly estimates camera parameters (extrinsics and intrinsics, when needed) and sparse or dense 3D point clouds by detecting and matching features across images, initializing a projective reconstruction, and refining it via bundle adjustment to minimize reprojection error. The process typically includes outlier rejection (e.g., RANSAC-based epipolar filtering), incremental or global registration of views, and optional integration with multi-view stereo for dense surface reconstruction, enabling metrically accurate scene models under appropriate calibration and scale constraints.