Back to publications page

CVPR 2011

Structure from motion for scenes with large duplicate structures

Richard Roberts
Georgia Institute of Technology,
Atlanta

   

Sudipta N. Sinha
Microsoft Research,
Redmond

   

Richard Szeliski
Microsoft Research,
Redmond

   

Drew Steedly
Microsoft,
Redmond


Overview

(a)
Duplicate structures present in the scene.
(b)
A standard structure from motion (Sfm) pipeline folds both instances into one in reconstruction.
(c)
Our method infers the erroneous inter-instance matches (shown in red in the match graph adjacency matrix) and discards them.
(d)
This produces an accurate reconstruction.

Abstract
Most existing structure from motion (SFM) approaches for unordered images cannot handle multiple instances of the same structure in the scene. When image pairs containing different instances are matched based on visual similarity, the pairwise geometric relations as well as the correspondences inferred from such pairs are erroneous, which can lead to catastrophic failures in the reconstruction.
    In this paper, we investigate the geometric ambiguities caused by the presence of repeated or duplicate structures and show that to disambiguate between multiple hypotheses requires more than pure geometric reasoning. We couple an expectation maximization (EM)-based algorithm that estimates camera poses and identifies the false match-pairs with an efficient sampling method to discover plausible data association hypotheses. The sampling method is informed by geometric and image-based cues. Our algorithm usually recovers the correct data association, even in the presence of large numbers of false pairwise matches.

datasets
Datasets
The image sequences are now available for download.
group1.zip (bldg, street)
group2.zip (books, cereal, cup)
group3.zip (desk, oats) 

@inproceedings {ROBERTS-CVPR11,
    key = "Roberts",
    author = "Richard Roberts and Sudipta N. Sinha and
              Richard Szeliski and Drew Steedly",
    title = "Structure from motion for scenes with large
             duplicate structures ",
    booktitle = "IEEE Computer Society Conference on Computer
                 Vision and Pattern Recognition (CVPR 2011)",
    location = "Fort Collins, CO",
    month = "June",
    year = "2011",
    pubid = "147401",
    keywords = "structure from motion, duplicate structures",
    bibdate = "Wed Apr  6 17:34:27  2011"
}
         
paper
 paper
supplementary page
 Supplementary Material
Back to publications page