Robust and Accurate Multi-View Reconstruction by Prioritized Matching

Markus Ylimäki, Juho Kannala, Jukka Holappa, Janne Heikkilä
Center for Machine Vision Research
University of Oulu

Sami S. Brandt
University of Copenhagen

The paper was published in the 21st International Conference on Pattern Recognition in Tsukuba, Japan.


This paper proposes a prioritized matching approach for finding corresponding points in multiple calibrated images for multi-view stereo reconstruction. The approach takes a sparse set of seed matches between pairs of views as input and then propagates the seeds to neighboring regions by using a prioritized matching method which expands the most promising seeds first. The output of the method is a threedimensional point cloud. Unlike previous correspondence growing approaches our method allows to use the best-first matching principle in the generic multi-view stereo setting with arbitrary number of input images. Our experiments show that matching the most promising seeds first provides very robust point cloud reconstructions efficiently with just a single expansion step. A comparison to the current state-of-the-art shows that our method produces reconstructions of similar quality but significantly faster.



Demo video below shows how the method works with three images.

Markus Ylimäki