Modeling Human Communication Dynamics for Virtual Human - Coverbal Synchrony in Human-Machine Interaction - page 368

Figure 3. A simplifi ed hierarchy for synthesizing motion. The rectangles indicate the parts of
the body controlled during each state, starting from the larger rectangles and going inward
towards the smaller rectangles. The entire body motion is synthesized for, in this case, a
sitting virtual human. Next, the upper body gesturing is synthesized by layering gesture
movements on top of the lower body. Next, spine movements controlling posture and gaze
are layered on top of the gesture. Then, head movements are added for backchannelling and
movement during speaking. Next, facial movements to express emotion and coordinate lip
movement, then eye movement including saccades and eyelid positioning.
(Color image of this fi gure appears in the color plate section at the end of the topic.)
have three small shakes of the hand, while the other only has two. In
addition, large movements across the body or broad variations in poses
across the gestures blend together poorly. Blended poses can vary in
completion time; it is unlikely that any two motion-captured gestures
will take the same amount of time. Gestures are blended together by
first timewarping, that is stretching or compressing, the motions to
the desired time, and then combining the various motions together.
A large difference in time between any two motions will lead to poor
quality blends, since one or both motions will need to be lengthened
or shortened to match the other, typically changing the dynamics of
motion that are embedded within the original captured motion. Thus,
a gesture that is synthesized from one or more blends maintains the
highest level of fidelity when the individual blended gestures have
matching phases, and similar timings. There are many different ways to
blend motions together, offering various trade-offs between execution
time and memory (Kovar and Gleischer, 2004; Huang and Kallmann,
2010) as well as tradeoffs between precision and smoothness (Pettre
and Laumond, 2006; Rose et al., 1998). An overview of common
blending techniques can be found in Feng et al. (2012).
2.5 Hierarchical gesture models
One model for achieving variations in gestures is to use a model of
hierarchical of control over the virtual human movement (Kallmann