Semi-Global 3D Line Modeling for Incremental Structure-from-Motion

Manuel Hofer, Michael Donoser and Horst Bischof

In Proceedings British Machine Vision Conference 2014


Structure-from-Motion (SfM) approaches, which are conventionally based on local interest point matches, tend to work well for richly textured indoor- and outdoor environments. However, in less textured scene areas the density of the resulting point cloud suffers from the lower number of matchable interest points. This significantly affects subsequent computer vision tasks like image based localization, surface extraction or visual navigation. In this paper, we propose a novel 3D reconstruction approach that increases the amount of 3D information in the reconstruction by exploiting line segments as complementary features. We introduce an efficient and effective semi-global approach, which takes into account local (per 2D line segment) as well as global (graph clustering) 3D line hypotheses constellations. Our approach outperforms the state-of-the-art in terms of accuracy, with comparable runtime.


Poster Session


Extended Abstract (PDF, 1 page, 1.3M)
Paper (PDF, 12 pages, 4.4M)
Bibtex File


Manuel Hofer, Michael Donoser, and Horst Bischof. Semi-Global 3D Line Modeling for Incremental Structure-from-Motion. Proceedings of the British Machine Vision Conference. BMVA Press, September 2014.


	title = {Semi-Global 3D Line Modeling for Incremental Structure-from-Motion},
	author = {Hofer, Manuel and Donoser, Michael and Bischof, Horst},
	year = {2014},
	booktitle = {Proceedings of the British Machine Vision Conference},
	publisher = {BMVA Press},
	editors = {Valstar, Michel and French, Andrew and Pridmore, Tony}
	doi = { }