Automated Cinematography

Abstract

Most efforts in motion capture have focused on low- or middle-level tasks such as data acquisition and editing. Higher level efforts have included reusing data for different figures and retargetting data to dynamically changing figures. However, relatively little effort has focused on these and other high-level tasks. One task is that of automating cinematography; such a process would be analogous to creating a digital director in place of a human one. In this project, I will attempt to automate some aspects (which are as yet undetermined) of cinematography by controlling the behavior of the camera in a way that produces meaningful scenes. The challenge will be inferring what is "meaningful" from a set of data about which nothing is known. Because not everyone will agree on what is meaningful or appealing, I will use the techniques employed by filmmakers and animators as a set of guidelines. Finally, although the goal of automation is not necessarily to eliminate the director, I will limit the scope of this project by producing scenes that cannot be edited.

Approach

I plan to use numerical measures of the data to try to understand the motion over various ranges of time. These measures will supply information such as frequency changes and the spatial arrangement and scope of the figure’s movements. From this data I can make decisions about the placement of the camera.

Other than this, my direction is unclear. I will update this page as I figure things out.