Key points are not available for this paper at this time.
We present a new environment for the development of active vision algorithms that are situated in a behavioral context. Our environment allows an unencumbered person to interact with a simulated graphical world which contains autonomous agents. Active vision techniques are used to observe the person and provide perceptual stimuli to the agents. An image of the person is composited together with the graphical world and projected onto a large video screen which the person can observe. The agents inhabiting the world are modeled as autonomous entities which have their own simulated sensors and goals and which can interpret the actions of the participant and react to them in real-time. The goals of these agents provide behavioral context for the execution of the low-level vision routines which observe the person. Several prototypical worlds have been implemented in this environment and tested on hundreds of users at a large-scale public installation.
Darrell et al. (Tue,) studied this question.