Information Technology Reference
In-Depth Information
Speeding Up Structure from Motion on Large
Scenes Using Parallelizable Partitions
Koen Douterloigne, Sidharta Gautama, and Wilfried Philips
Department of Telecommunications and Information Processing
(UGent-TELIN-IPI-IBBT)
Ghent University, St-Pietersnieuwstraat 41, B-9000 Ghent, Belgium
koen.douterloigne@telin.ugent.be
Abstract. Structure from motion based 3D reconstruction takes a lot
of time for large scenes which consist of thousands of input images. We
propose a method that speeds up the reconstruction of large scenes by
partitioning it into smaller scenes, and then recombining those. The
main benefit here is that each subscene can be optimized in parallel.
We present a widely usable subdivision method, and show that the dif-
ference between the result after partitioning and recombination, and the
state of the art structure from motion reconstruction on the entire scene,
is negligible.
Keywords: Structure from motion, 3D reconstruction, speedup, large
scenes.
1
Introduction
More and more applications require accurate three dimensional (3D) models
of the world, e.g. planning urban environments and infrastructures, automated
object detection, or augmented reality and CGI in movies. To create these 3D
models various options exist, including mobile mapping, laser scanning, or man-
ual surveying. These ground based acquisition methods all have in common that
they are time consuming, especially for larger areas. The most practical approach
to quickly cover a lot of terrain is that of aerial imaging, where 2D pictures are
captured and then processed to create a 3D model [4]. Just like in [13], the 3D
model is derived from multiple pictures of the same area, taken under different
angles.
To find the initial position of the pictures usually GPS information is used.
A major problem however is that we can not always rely on GPS information
being available. In land based mapping anything that interferes with the line of
sight to the GPS satellite has a negative effect on the reception, e.g. trees or
large buildings. But even low altitude aerial surveillance can not always count
on a GPS link, as certain regions have active jamming devices (e.g. conflict
zones). A good solution to handle these problems is to employ structure from
motion. Instead of relying on the GPS information, the position of the camera
is determined from image correspondences. Consecutive aerial pictures have a
 
Search WWH ::




Custom Search