Three-Dimensional Pose Estimation and Segmentation Methods - 3D Computer Vision: Efficient Methods and Applications - page 95

Fig. 2.2 Matching results
(best solution) for several
example poses
for an automotive part in Fig. 2.2 . A quantitative evaluation of the described edge-
based pose estimation technique is performed in the scenario of industrial quality
inspection in Sect. 6.1.1 .
To improve the robustness of monocular pose estimation in the presence of a
cluttered background, the edge-based method is extended by Barrois and Wöhler
( 2007 ) to appearance-based monocular pose estimation based on geometric, pho-
topolarimetric, and defocus cues (cf. Sect. 5.6 ).
2.2 Pose Estimation of Non-rigid and Articulated Objects
In contrast to rigid objects, articulated objects consist of several rigid subparts which
are able to move with respect to each other. Methods that aim for a pose estimation
of such objects need to determine these internal degrees of freedom in addition to
the six rotational and translational degrees of freedom encountered for rigid ob-
jects. Non-rigid objects have no rigid subparts at all and therefore have an infinite
number of internal degrees of freedom. In this section we first give an overview of
pose estimation methods for articulated and non-rigid objects. Then we regard the
contour-based algorithm of d'Angelo et al. ( 2004 ) for three-dimensional reconstruc-
tion of non-rigid objects such as tubes and cables, which may be regarded as a mul-
tiocular extension of the concept of active contours (Blake and Isard, 1998 ). Sub-
sequently, the multiocular contracting curve density algorithm (Hahn et al., 2007 ,
2010a ) is described, which allows a three-dimensional reconstruction of non-rigid
objects and a pose estimation of articulated objects in the presence of a cluttered
background.
2.2.1 General Overview
2.2.1.1 Non-rigid Objects
The extraction of two-dimensional object contours from images is an essential part
of image segmentation. A classical approach to this problem is that of active con-
tours or snakes. The original snake algorithm by Kass et al. ( 1988 ) determines a
curve for which an optimisation is performed simultaneously in terms of its length
or curvature and its correspondence with edges in the image. Many variations and