Triangulation-Based Approaches to Three-Dimensional Scene Reconstruction - 3D Computer Vision: Efficient Methods and Applications - page 73

An important advantage of the spacetime stereo method described in this section
is the fact that no explicit correspondences need to be established over time. Further-
more, the motion parameters are available nearly instantaneously (after acquisition
of three images in the presented examples) since no tracking stage is involved—
tracking systems usually require a certain settlement phase after initialisation before
the estimated motion parameters become reliable. In Chap. 7 it is demonstrated that
spacetime stereo is a useful technique in the context of three-dimensional scene seg-
mentation and object tracking. Especially in the presence of several objects in the
scene which all move in a different manner, it is often difficult to assign parts of the
three-dimensional point cloud to specific objects when only spatial information is
available. Adding motion cues to the point cloud introduces new information that
may allow one to distinguish unambiguously between the objects in the scene even
when they come close to each other or mutually overlap.
1.6 Resolving Stereo Matching Errors due to Repetitive
Structures Using Model Information
It is a 'universal problem' of stereo vision that repetitive structures may lead to
spurious objects in front of or behind the true scene (cf. Fig. 1.24 ), which may
cause severe problems in scenarios involving mobile robot navigation or human-
robot interaction. To alleviate this problem, a model-based method is proposed by
Barrois et al. ( 2010 ) which is independent of the specific stereo algorithm used.
The basic idea is the feedback of application-dependent model information into the
correspondence analysis procedure without losing the ability to reconstruct scene
parts not described by the model. The description in this section is adopted from
Barrois et al. ( 2010 ).
Many stereo algorithms attempt to avoid false correspondences by using well-
known techniques such as the ordering constraint, the smoothness constraint, the
geometric similarity constraint, or a left-right consistency check (Fua, 1993 ). Re-
garding repetitive structures, Di Stefano et al. ( 2004 ) assess the quality of the mini-
mum of the cost function and the related disparity value by introducing a distinctive-
ness and a sharpness test to resolve ambiguities. Some approaches handle erroneous
stereo correspondences explicitly. Murray and Little ( 2004 ) use the RANSAC algo-
rithm (Fischler and Bolles, 1981 ) to fit planes to the three-dimensional points in or-
der to detect and eliminate gross errors. Sepehri et al. ( 2004 ) use a similar approach
to fit a plane to the three-dimensional points of an object using an M-estimator tech-
nique (Rey, 1983 ).
Barrois et al. ( 2010 ) present a novel method to cope with repetitive structures
in stereo analysis, which can be applied independent of the specific stereo algo-
rithm used. In a first step, a three-dimensional reconstruction of the scene is de-
termined by conventional correspondence analysis, leading to correct and incorrect
three-dimensional points. An application-dependent scene model or object model
is adapted to the initial three-dimensional points, which yields a model pose. The