Automated cinematography and editing for 3D computer graphics scenes

Abstract

We present an approach to generate cinematic sequences from event-based descriptions of 3D computer graphics scenes of conversations between groups of actors. We break the problem into two parts: cinematography and editing. Our cinematography approach creates camera setups using a combination of geometric constraints and aesthetic parameters, while ensuring that the resulting cinematic sequence obeys the heuristics of traditional cinematography. More specifically, our main contributions are a method for defining hierarchical lines of action and the identification and use of relevant first principles of cinematography for using these lines of actions. Our approach is more flexible and powerful than those proposed in previous work, mainly because it naturally generalizes to any number of actors in a scene. On the editing side, we present a set of heuristics for editing shots into a coherent movie clip which obeys the conventions of continuity editing. Our approach mimics the decision processes of an editor assembling a clip out of filmed footage involving multiple camera setups. Given a set of stylistic rules, our software applies a number of heuristics to produce a final result satisfying those rules, as well as the fundamental rules of continuity editing. The main contribution of this work is in the formulation of editing heuristics which take into account stylistic rules, enabling different edits of the same scene into cinematic clips of various styles. We demonstrate the use of these heuristics on three scenes taken from actual film clips.