Introduction - Multimedia Database Retrieval: Technology and Applications - page 13

Fig. 1.5 3D motion database retrieval in a virtual reality dance training system

graph cut for video object segmentation. It is particularly effective when applied
to video objects appearing at weak edges, with poor lamination distribution, and
having backgrounds of similar color and movement. In addition, human faces are
also interesting video objects. Detection algorithms have to cope with inconsistent
performance due to sensitivity to lamination variations such as local shadowing,
noise, and occlusion. The face detection method overcomes these problems by the
incorporation of the local histogram and optimal adaptive correlation methods.

1.3.5.7 3D Motion Database Retrieval

3D motion data is acquired by the sensors in a process which is different from the
acquisition of image and video data. The 3D motion data of human movement is
represented by a time series of the body's joint positions. It can be captured by
Microsoft Kinect in terms of the skeleton tracking of joint positions, as well as
by the full motion capture system using optical-reflective markers. The usefulness
of 3D motion database applications can be seen in the recent trend towards more
immersive and interactive computing. This application requires tools to analyze,
understand, and interpret human motion, in particular human gestural input. Human
gestures include movement of the hands, arms, head, face or body with the intention
of conveying meaningful information or interacting with the environment.
Figure 1.5 show the application of 3D motion database retrieval in a dance
training system. In a virtual reality dance training system, dance gesture recognition
is the key issue in the comparison of captured dance motion taken in real time
from the trainee against the trainer data. The recognition result will enable the
execution of subsequent tasks, such as automatic dance performance assessment,
and synthesizing virtual dance characters and dance partners in the VR settings.
In Chap. 11 , a dance training system is presented for automatic dance gesture
recognition. The system adopts the spherical self-organizing map (SSOM) for the
unsupervised parsing of dance movement into a structured posture space, which