Estimating collective attention toward a public display

Enticing groups of passers-by to focused interaction with a public display requires the display system to take appropriate action that depends on how much attention the group is already paying to the display. In the design of such a system, we might want to present the content so that it indicates that a part of the group that is looking head-on at the display has already been registered and is addressed individually, whereas it simultaneously emits a strong audio signal that makes the inattentive rest of the group turn toward it. The challenge here is to define and delimit adequate mixed attention states for groups of people, allowing for classifying collective attention based on inhomogeneous variants of individual attention, i.e., where some group members might be highly attentive, others even interacting with the public display, and some unperceptive. In this article, we present a model for estimating collective human attention toward a public display and investigate technical methods for practical implementation that employs measurement of physical expressive features of people appearing within the display's field of view (i.e., the basis for deriving a person's attention). We delineate strengths and weaknesses and prove the potentials of our model by experimentally exerting influence on the attention of groups of passers-by in a public gaming scenario.