Parallel Generalized Thresholding Scheme for Live Dense Geometry from a Handheld Camera.

Stühmer, Jan; Gumhold, Stefan; Cremers, Daniel

doi:10.1007/978-3-642-35740-4_35

by Jan Stühmer, Stefan Gumhold, Daniel Cremers

Abstract:

Inspired by recent successes in parallelized optic flow estimation, we propose a variational method which allows to directly estimate dense depth fields from a single hand-held camera in real-time conditions. In particular we show how the central ingredient of the corresponding optic flow method, namely a thresholding scheme, can be generalized to the problem of geometric reconstruction considered in this paper and how it can be parallelized on recent graphics cards. We compare alternative parallelization strategies and experimentally validate that high-quality depth maps can be computed in a few milliseconds from a hand-held camera.

View PDF

Reference:

Parallel Generalized Thresholding Scheme for Live Dense Geometry from a Handheld Camera. (Jan Stühmer, Stefan Gumhold, Daniel Cremers), In ECCV Workshops (1), volume 6554, 2010.

Bibtex Entry:

@inproceedings{Stuhmer-2010-PGT,
  author = {St\"uhmer, Jan and Gumhold, Stefan and Cremers, Daniel},
  booktitle = {ECCV Workshops (1)},
  affiliations = {CGV},
  areas = {areasa,areasu},
  pages = {450-462},
  series = {Lecture Notes in Computer Science},
  title = {Parallel Generalized Thresholding Scheme for Live Dense Geometry from a Handheld Camera.},
  volume = {6554},
  year = {2010},
  keywords = {3d-reconstruction},
  abstract = {Inspired by recent successes in parallelized optic flow estimation, we propose a variational method which allows to directly estimate dense depth fields from a single hand-held camera in real-time conditions. In particular we show how the central ingredient of the corresponding optic flow method, namely a thresholding scheme, can be generalized to the problem of geometric reconstruction considered in this paper and how it can be parallelized on recent graphics cards. We compare alternative parallelization strategies and experimentally validate that high-quality depth maps can be computed in a few milliseconds from a hand-held camera.},
  doi = {10.1007/978-3-642-35740-4_35},
  url = {http://link.springer.com/chapter/10.1007%2F978-3-642-35740-4_35}
}