Three-Dimensional Pose Estimation and Segmentation Methods - 3D Computer Vision: Efficient Methods and Applications - page 87

Chapter 2
Three-Dimensional Pose Estimation
and Segmentation Methods
In the previous chapter it has been described how a three-dimensional reconstruction
of a scene can be obtained from point correspondences between images, and how
this problem is related with the problem of determining the intrinsic and extrinsic
camera parameters. In this context, the reconstruction result is always obtained as
a cloud of three-dimensional points. This point cloud contains information about
the presence of matter at a certain position in space. However, no information is
available about the presence or even the position and orientation of objects in the
scene.
Hence, this chapter provides an analysis of traditional and newly introduced
methods for the segmentation of point clouds and for the three-dimensional detec-
tion and pose estimation of rigid, articulated, and flexible objects in a scene based
on single or multiple calibrated images, where it is assumed that a (more or less
detailed) model of the object is available.
For rigid objects, pose estimation refers to the estimation of the six degrees of
freedom of the object, i.e. three rotation angles and three translation components
(cf. Sect. 2.1 ). For objects with a limited number of internal degrees of freedom
('articulated objects') and objects with an infinite number of internal degrees of
freedom ('non-rigid objects') pose estimation may correspond to a much higher
dimensional optimisation problem (cf. Sect. 2.2.1.2 ). Here, the three-dimensional
object detection is based directly on the acquired images of the scene or suitable
features derived from them. A class of complementary approaches deals with the
analysis of point clouds obtained e.g. by stereo image analysis with respect to their
three-dimensional structure and motion behaviour, aiming for a segmentation into
physically meaningful domains representing individual objects or object parts. Such
methods are regarded in Sect. 2.3 .
2.1 Pose Estimation of Rigid Objects
To estimate the pose of a rigid object, a set of three-dimensional scene points
with coordinates given by a geometry model of the object is brought into corre-