Synthesis and Analysis Techniques for the Human Body: R&D Projects - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 342

synthesis and analysis techniques for the human body reported by R&D
projects worldwide. Technical details are provided for each R&D project
and the results are discussed and evaluated.
Introduction
Humans are the most commonly seen moving objects in one's daily life. The
ability to model and to recognize humans and their activities by vision is key for
a machine to interact intelligently and effortlessly with a human inhabited
environment. Because of many potentially important applications, examining
human body behavior is currently one of the most active application domains in
computer vision. This survey identifies a number of promising applications and
provides an overview of recent developments in this domain (Hillis, 2002).
Hand and body modeling and animation is still an open issue in the computer
vision area. Various approaches to estimate hand gestures and body posture or
motion from video images have been previously proposed (Rehg & Kanade,
1994; Lien & Huang, 1998; Zaharia, Preda & Preteux, 1999). Most of these
techniques rely on 2-D or 3-D models (Saito, Watanabe & Ozawa, 1999; Tian,
Kanade & Cohn, 2000; Gavrila & Davies, 1996; Wren, Azarbayejani, Darell &
Pentland, 1997) to compactly describe the degrees of freedom of hand and body
motion that has to be estimated. Most techniques use as input an intensity/color
image provided by a camera and rely on the detection of skin color to detect
useful features and to identify each body part in the image (Wren, Azarbayejani,
Darell & Pentland, 1997). In addition, the issue of hand and body modeling and
animation has been addressed by the Synthetic/Natural Hybrid Coding (SNHC)
subgroup of the MPEG-4 standardization group to be described in more detail in
the following.
In Sullivan & Carlsson (2002), view-based activity recognition serves as an input
to a human body location tracker with the ultimate goal of 3D reanimation. The
authors demonstrate that specific human actions can be detected from single
frame postures in a video sequence. By recognizing the image of a person's
posture as corresponding to a particular key frame from a set of stored key
frames, it is possible to map body locations from the key frames to actual frames
using a shape-matching algorithm. The algorithm is based on qualitative similarity
that computes point-to-point correspondence between shapes, together with
information about appearance.
In Sidenbladh, Black & Sigal (2002), a probabilistic approach is proposed to
address the problem of 3D human motion modeling for synthesis and tracking.
High dimensionality and non-linearity of human body movement modeling is