Discovering Human Postures from Non-human Silhouette

Abstract

The shadow theatre is an unique performing arts, which utilizes a shadow as an main communication tool. This can be understood easily by an extension of traditional shadow play. Key difference comes from the fact that entire human bodies are used to make interesting shapes on screen during the play, whereas puppets or human hands are only used in the traditional one. Modern technologies on projection devices facilitate larger stage area, which enables the actors to use human motions in the play. Viewers often prefer full body shadow play over the traditional one because it gives energetic and delicate feelings simultaneously and it is less trivial to conjecture original performers' postures. Our goal is to automatically discover human postures from nonhuman silhouettes or shadow images. We are particularly interested in the situation where multiple actors and environments(e.g., floor, platform or props) exist similar to the true performance. The actors' posing process is therefore greatly accelerated by our system for new shapes which haven't been handled yet. The system can also be used to generate articulated sculptures imitating target shadow whereas freeform geometry was made in [Mitra and Pauly 2009]. A light source, screen and actors are substructional elements in the shadow theatre. If their positions and orientations are fixed, one scene is determined accordingly. Precise and complex coordination between the elements in spatial domain is therefore key to choreograph. To do so, we present a novel approach that takes as input in the form of non-human 2D silhouettes or any types of images which can represent target shadows. From this input our system discovers a set of human poses.