Integrated Frameworks for Three-Dimensional Scene Reconstruction - 3D Computer Vision: Efficient Methods and Applications - page 186

but a sufficient amount of surface texture is required. Due to the fact that estimation
of the PSF is sensitive with respect to pixel noise, the resulting depth values tend
to be rather inaccurate. Depth from focus is very accurate but also time-consuming
due to the large number of images required; thus it should be preferentially applied
in measurement applications rather than computer vision scenarios.
These considerations illustrate that each of the approaches described in
Chaps. 1 - 4 has its specific advantages and drawbacks. Some of the techniques
are complementary; as an example, triangulation-based methods yield three-
dimensional point clouds describing textured parts of the scene while intensity-
based methods may be able to reconstruct textureless regions between the points.
Hence, just as the human visual system achieves a dense three-dimensional scene re-
construction based on combinations of different cues, it appears to be favourable for
computer vision systems to integrate different three-dimensional scene reconstruc-
tion methods into a unifying framework. This chapter describes several approaches
of this kind and discusses their specific preconditions, advantages, limitations, and
preferential application domains.
5.1 Monocular Three-Dimensional Scene Reconstruction
at Absolute Scale
This section describes a method for combining triangulation-based and PSF-based
methods for monocular three-dimensional reconstruction of static scenes at absolute
scale, as introduced by Kuhl et al. ( 2006 ) and Wöhler et al. ( 2009 ). The presentation
in this section is adopted from Wöhler et al. ( 2009 ).
The described algorithm relies on a sequence of images of the object acquired by
a monocular camera of fixed focal setting from different viewpoints. Object fea-
tures are tracked over a range of distances from the camera with a small depth
of field, leading to a varying degree of defocus for each feature. Information on
absolute depth is obtained based on a depth from defocus approach. The parame-
ters of the PSFs estimated by depth from defocus are used as a regularisation term
for structure from motion. The reprojection error obtained from bundle adjustment
and the absolute depth error obtained from depth from defocus are simultaneously
minimised for all tracked object features. The proposed method yields absolutely
scaled three-dimensional coordinates of the scene points without any prior knowl-
edge about scene structure and the camera motion. The implementation of the pro-
posed method is described both as an offline and as an online algorithm. Evaluating
the algorithm on real-world data, we demonstrate that it yields typical relative scale
errors of a few percent. We examine the influence of random effects, i.e. the noise
of the pixel grey values, and systematic effects, caused by thermal expansion of the
optical system or by inclusion of strongly blurred images, on the accuracy of the
three-dimensional reconstruction result.
To our knowledge, prior to the work by Kuhl et al. ( 2006 ) no attempt has been
made to combine the precise relative scene reconstruction of structure from motion