Graphics Reference
In-Depth Information
and orientation of the joint changes. Thus, 3D characters that consist
of a 3D mesh can move, stretch and twist such a mesh in a way that
appears somewhat natural by simply rotating and translating the joints
in the underlying skeleton. Thus, it is often sufficient to generate the
motion for a 3D character in order to appropriately express a gesture
or movement when using these common 3D character animation
techniques. Thus, the challenge of synthesizing gestures is generally
associated with acquiring and synthesizing skeletal motion.
Other methods of generating digital characters and motions exist,
such as the use of image-based techniques, where an entire mesh,
texture and movement are captured at the same time. However, at
the time of this writing, such methods are not mature and are not
widely supported.
There are several different kinds of gesture and expression
architectures that can be used, each offering a different level of
quality, different data requirements, different flexibility and varying
complexity of use. In general, the gesture architectures that offer the
highest level of quality are those that use motion capture data explicitly
and replay it without modification. Those with the lowest level of
quality segment gestures into smaller phases then synthesize motions
procedurally from various algorithms. Gesture architectures with the
greatest level of control and precision generally favor the reverse;
procedurally generated, phased gesture motion provides better control
than replayed motion capture clips. The following sections describe
variations in a gesture architecture based on these ideas.
2.1 Motion capture session
In general, a motion capture session requires a motion capture system
based on cameras or inertial sensors. The session typically requires
transferring the data onto a virtual character during the capture
session. The motion capture process typically synthesizes data onto
an existing skeleton, which does not match exactly the proportions
and sizes of the real human actor. Thus, the captured data needs to be
retargeted or transferred onto the skeleton which models the virtual
character. Motion capture data is typically segmented into clips,
each clip representing semantically related content. Similarly, longer
motion capture clips can be segmented into smaller ones during a
post-processing phase, where data is refined and edited. The specifics
of capturing motion via motion capture can be found in other sources.
Below we describe several capture and motion synthesis strategies.
Search WWH ::
Custom Search