A Semantic Space Partitioning Approach to Virtual Camera Composition

In this paper, we present a semantic space partitioning (SSP) approach to the virtual camera composition problem. Virtual camera composition (VCC) consists in positioning a camera in a virtual world, such that the resulting image satisfies a set of visual cinematographic properties. Whereas most related works concentrate on numerically computing a unique camera position satisfying the problem, we offer to isolate identical possible solutions in 3D volumes with respect to their visual properties, and to propose them to the user. We introduce the notion of semantic volumes as an extension of visual aspects to characterize, compute and manipulate distinct solution sets. Our approach relies on (1) a space partitioning process derived from a study of possible camera locations w.r.t. to the objects in the scene and (2) local search numerical techniques to compute good representatives of each volume. This work is motivated by the lack of VCC tools in 3D software and the will to integrate cinematographic semantics in the description, solving and interaction processes. Experimental results illustrate the suitability of our approach for identifying and providing distinct solution sets. Furthermore, the exploitation of the semantic volumes lays the groundwork for natural and efficient user interaction by providing knowledge and reasoning on possible classes of solutions.