Key points are not available for this paper at this time.
This article describes strategies to model the geometry and actions occurring in a performance space such as a TV studio. The basic assumption is that in such situations a script is available, providing contextual information that can be coupled with visual data from cameras monitoring the space. The integration of linguistic and visual information is accomplished through the use of approximate world models, a combination of symbolic data and geometrical 3-D models with low positional accuracy. Some commonsense reasoning is often required to infer visual and geometrical attributes from the script, and the article examines how simple inference rules are able to provide most of the necessary information. We also detail an application where an intelligent studio controls three automatic robotic cameras, producing video sequences with TV quality of framing.
Pinhanez et al. (Sun,) studied this question.