Real-Time Analysis of Human Body Parts and Gesture-Activity Recognition in 3D - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 147

Figure 3. Algorithm blocks and corresponding results of selected steps.
bine the meaningful adjacent segments and use them as the input of the
following algorithm steps.
Contour following: We apply the contour following algorithm that uses the 3x3
filter to follow the edge of the component where the filter can move in any
of eight directions to follow the edge.
Ellipse fitting: Even when human body is not occluded by another object, due
to the possible positions of non-rigid parts, a body part can be occluded in
different ways. For example, the hand can occlude some part of the torso
or legs. In this case, 2D approximation of parts by fitting ellipses with shape-
preserving deformations provides more satisfactory results. It also helps to
discard the deformations due to the clothing. Global approximation methods
give more satisfactory results for human detection purposes. Hence,
instead of region pixels, parametric surface approximations are used to
compute shape descriptors. Details of the ellipse fitting can be found in
Ozer & Wolf (2002b).
Object modeling by invariant shape attributes: For object detection, it is
necessary to select part attributes which are invariant to two-dimensional
transformations and are maximally discriminating between objects. Geo-
metric descriptors for simple object segments such as area, compactness
(circularity), weak perspective invariants, and spatial relationships are
computed (Ozer et al., 2000). These descriptors are classified into two
groups: unary and binary attributes. The unary features for human bodies
are: a) compactness, and b) eccentricity. The binary features are: a) ratio