With the goal of building an animated agent, Lücking et al.
[4] investigated SaGA corpus, which is a detailed and
systematically annotated speech-and-gesture corpus, and
categorized eight types of gestures: Indexing, Placing, Shaping,
Drawing, Posturing, Sizing, Counting, and Hedging. Because
we aim to generate iconic gestures that illustrate the shape of
the objects, we focus on drawing gestures: hands moving as if
tracing the outline of an object’s shape. As a study of gesture
generation, Sadeghipour et al. [5] investigated the structures of
iconic gesturing performed by human subjects in an
experimental setting, and extracted compositional patterns of
iconic gesturing. They then proposed Feature-based Stochastic
Context-Free Grammars (FSCFG), which can be used as
gesture generation rules. In this study, we do not use human
gesturing data, but extract characteristics of object shapes to be
illustrated in drawing gestures using image processing
techniques.