VocMatch: Efficient Multiview Correspondence for Structure from Motion

摘要: Feature matching between pairs of images is a main bottleneck structure-from-motion computation from large, unordered image sets. We propose an efficient way to establish point correspondences all in dataset, without having test each individual pair. The principal message this paper that, given sufficiently large visual vocabulary, feature can be cast as indexing, subject the additional constraints that index words must rare database and unique image. demonstrate proposed method, conjunction with standard inverted file, 2-3 orders magnitude faster than conventional pairwise matching. vocabulary-based has been integrated into SfM pipeline, delivers results similar those method much less time.