Efficient Structure from Motion by Graph Optimization

Abstract

Abstract. We present an efficient structure from motion algorithm that can deal with large image collections in a fraction of time and effort of previous approaches while providing comparable quality of the scene and camera reconstruction. First, we employ fast image indexing using large image vocabularies to measure visual overlap of images without running actual image matching. Then, we select a small subset from the set of input images by computing its approximate minimal connected dominating set by a fast polynomial algorithm. Finally, we use task prioritization to avoid spending too much time in a few difficult matching problems instead of exploring other easier options. Thus we avoid wasting time on image pairs with low chance of success and avoid matching of highly redundant images of landmarks. We present results for several challenging sets of thousands of perspective as well as omnidirectional images.

Selecting image pairs for SfM by introducing Jaccard Similarity

“…And those photos would cover large parts of the Earth. Several techniques utilizing photos on the Web for SfM have appeared [2][3][4][5][6][7].…”

Efficient Wide Baseline Structure From Motion

“…Recent developments for Structure from Motion (SfM) techniques from unorganized image sets focus on large photo collections downloaded from the internet (Heinly et al, 2015;Snavely et al, 2008;Agarwal et al, 2009;Frahm et al, 2009;Havlena et al, 2010;Crandall et al, 2011). Such collections can contain thousands or even millions of images comprising a very high redundancy and often moderate baselines.…”

Towards Fast Image-Based Localization on a City-Scale

“…An especially interesting application is image-based localization for mobile devices, where a user simply sends a photo taken with her mobile phone to a server and in return receives information about her position [7]. Camera positions computed by localization methods are also useful for Structure-from-Motion reconstructions [1,10,14,18,28,35] or for the visualization of photo collections [34].…”

“…Furthermore, many confusing features found in the images have to be removed for better retrieval [22]. The redundancy in the image set can be exploited to obtain a Structure-from-Motion (SfM) reconstruction of the scene [1,10,14,18,28,35], resulting in a 3D point cloud. The image matching part of the SfM pipeline automatically removes most of the confusing features.…”