Human and Virtual Agent Expressive Gesture Quality Analysis and Synthesis - Coverbal Synchrony in Human-Machine Interaction - page 278

these animated gestures are evaluated in a perceptive study. Third,
in order to find the relation between perceived emotion and the low-
level features of evaluated gestures' animations, the authors apply
rough set exploration algorithms on the low-level features of animated
gestures (i.e., the amplitude, the length and the speed of every joint's
movement). The authors show that the most significant features for
communicating emotional content are related to the duration of the
gesture phases (stroke position, duration of hold, preparation and
stroke phase). They argue that the duration of the gesture phases is
the most significant feature in communicating emotional states; it is
more important than the amplitude of the movement. Finally they
also propose values of low-level gesture features that can be applied
to any gesture animation in order to modify its emotional content.
Hsieh and Luciani (2005, 2006) use a physically based particle
modeling approach to model several modern dance figures. Their
models allow the user to control the expressive qualities of movement
such as light vs. strong, free vs. bound, or sudden vs. sustained. In
more details, for each dance figure the authors define one physically
based particle model. Each figure needs its own description using
different set of parameters (e.g., masses, forces) that influence the
dynamics of the movement. Thus, every dance figure is defined using
different set of mathematical equations that describe its forces, velocity,
potential, and kinetic energy. The models describe the energy flow
for each figure rather than its spatial criteria. So, instead of defining
explicitly its trajectory, a movement is generated implicitly by the
involved forces. The single models can be combined in more complex
behaviors.
2.3 Expressive gesture quality loop between
humans and machines
Applications offering real-time expressive gesture quality analysis and
synthesis are becoming widely developed in multimodal interactive
scenarios. Their final aim is the creation of credible and natural
multimodal interaction between human and machine: that is, they aim
to create an expressive gesture quality loop , i.e., bidirectional interaction
that exploits the communicative role of gesture expressivity, as
illustrated in Figure 1.
The idea of an interactive HCI system that intentionally uses the
expressive qualities of the behavior was proposed by Caridakis et al.
(2007). Their system allows a virtual agent to mimic the qualitative
features of the user's behavior. For this purpose, recognized expressive