Human and Virtual Agent Expressive Gesture Quality Analysis and Synthesis - Coverbal Synchrony in Human-Machine Interaction - page 271

their emotional state, personality, or social role. The ability for systems
to understand users' behavior and to respond to them with appropriate
feedback is an important requirement for generating socially tuned
machines (Schröder et al., 2011; Urbain et al., 2010). Indeed, the
expressive gesture quality of movement is a key element in both
understanding and responding to users' behavior.
Many researchers (Johansson, 1973; Wallbott and Scherer, 1986;
Gallaher, 1992; Ball and Breese, 2000; Pollick, 2004) investigated human
motion features and encoded them into categories. Some authors refer
to body motion using dual qualifiers such as slow/fast, small/large,
weak/energetic, and unpleasant/pleasant. Behavior expressivity has
been correlated to energy in communication, to the relation between
temporal/spatial features of gestures, and/or to personality/emotion.
Harald G. Wallbott (Wallbott, 1998) deems that behavior expressivity
is related to the notion of quality of the mental, emotional, and/or
physical state, and the intensity of this state. Behaviors do not only
encode content information, that is, “ What is communicated ” through
a gesture shape, but also expressive information, that is, “ How it is
communicated ” through the manner of execution of the gesture.
There exist at least two important aspects of expressive gesture
quality analysis, that is, the low level feature detection and its high
level interpretation in terms of its eventual communicative meaning.
Both of them have received important contributions in the last years.
In next two subsections, we present both these aspects.
2.1.1 Expressive gesture features detection
Several low-level features were proposed to describe the expressivity
of the movement. Theories from arts and humanities, such as for
example Laban's Effort theory (Laban and Lawrence, 1947) are some
of the sources analysis techniques are grounded on. Several algorithms
have been proposed to measure the features that can be extracted
from a movement. Interestingly, the same features can be computed
in many different ways.
The Spatial Extent and the Fluidity of movement are two such
features that are often analyzed. Among others, Cardakis et al. (2007)
analyze the user's gesture extent by measuring the distance between
two hands, whereas hands' fluidity is computed as the sum of the
variance of the norms of the hands' motion vectors. Similarly, Camurri
et al. (2004a) compute the Contraction Index , which is the ratio between
the area of the minimum rectangle surrounding the actor body and
the body silhouette. In Bernhardt and Robinson's (2007) work, the
maximum distance of hand and elbow from body is taken. Camurri et