Conditioning Gaze-Contingent Systems for the Real World: Insights from a Field Study in the Fast Food Industry

Abstract

Eye tracking can be used to infer what is relevant to a user, and adapt the content and appearance of an application to support the user in their current task. A prerequisite for integrating such adaptive user interfaces into public terminals is robust gaze estimation. Commercial eye trackers are highly accurate, but require prior person-specific calibration and a relatively stable head position. In this paper, we collect data from 26 authentic customers of a fast food restaurant while interacting with a total of 120 products on a self-order terminal. From our observations during the experiment and a qualitative analysis of the collected gaze data, we derive best practice approaches regarding the integration of eye tracking software into self-service systems. We evaluate several implicit calibration strategies that derive the user’s true focus of attention either from the context of the user interface, or from their interaction with the system. Our results show that the original gaze estimates can be visibly improved by taking into account both contextual and interaction-based information.