Human and Virtual Agent Expressive Gesture Quality Analysis and Synthesis - Coverbal Synchrony in Human-Machine Interaction - page 272

al. (2004a) approximate movement fluidity with the Directness Index ,
revealing whether a movement follows a straight line or a sinuous
trajectory. Mazzarino et al. (2007) propose two different methods
to measure fluidity. First, they estimate hand's fluency by finding
gesture start and ending time, then they determine the amount of
movement phases in a given time window: the lower the number of
phases, the higher the fluency. Second, they analyze discrepancies
between different body parts' movements: fluency is then evaluated
by comparing the quantity of motion of upper and lower body.
More recently, Mazzarino and Mancini (2009) propose to estimate
human movement Smoothness by computing the correlation between
trajectory curvature and velocity in a given time window. Smoothness
is computed for each frame of a video and for a particular point on
the body. Thus, it is possible to establish several points (e.g., the two
hands) and compute smoothness separately for each of them.
The other group of expressive characteristics of a gesture focuses
on different temporal aspects of its realization. For instance, temporal
aspects of gesture in Bernhardt and Robinson (2007) are measured by
the average hand (and elbow) speed. A slightly different method to
estimate the temporal quality of the gesture is used in Mancini and
Castellano (2007). For this purpose, the authors compute the velocity
of the barycenter of the hand.
Movement Power and Impulsivity were also addressed by different
computational methods. In Cardakis et al. (2007), power is the first
derivative of the motion vectors, whereas Mancini and Castellano
(2007) operationalize power by the acceleration of the hand. More
complex algorithms are used to compute impulsivity of the movement.
In Mazzarino and Mancini (2009), it is characterized as a local peak
in the time series of quantity of motion. For this purpose, the authors
detect any significant rise of quantity of motion in a given time
window.
Finally, the Overall Body Activation is analyzed by Camurri et al.
(2004a) through the computation of the Quantity of Motion ( QoM ). It
is measured as the difference of the person's body silhouettes area
computed on consecutive video frames. In Cardakis et al. (2007), the
user's movement is estimated as the sum of the motion vectors of
color-tracked user's hands.
2.1.2 Communicative meaning of expressive gesture qualities
The high-level meaning of expressive gesture features has been
recently investigated, to extract the communicative high-level message
of gestures and body movements. Camurri et al. (2003, 2006) and