Extending Morphs in AZee Using Pose Space Deformations

Signing avatars have become increasingly important for sign language synthesis. However, to behave realistically, they must be able to replicate the coordinated activity of human hand movements and facial expressions. Most methods currently evaluate such motion using just kinematic techniques, which can limit the realism of the virtual characters. We propose a new methodology for creating a set of morphs in the AZee language to address this issue. We encapsulate a set of human movements and map their respective pose space deformations within our morphs. This allows us to capture the rigid as well as the non-rigid shape changes of the human anatomy and also addresses the stretching and contracting of the skin at its extremities. We create our pose space deformations based on the study of local avatar movements and a popular cognitive facial model for facial expressions. We integrate our set of morphs in our existing blender addon implementation for AZee with a standard parameterized 3D avatar model, resulting in a fully articulated avatar that can produce more realistic movements with a faster real-time synthesis. The proposed methodology has the potential to enhance the realism of signing avatars and contributes to the development of a more intuitive toolkit for AZee linguists.