Detecting Gaze Direction Using Robot-Mounted and Mobile-Device Cameras

Abstract

Two common channels through which humans communicate are speech andgaze. Eye gaze is an important mode of communication: it allows people tobetter understand each others’ intentions, desires, interests, and so on. The goalof this research is to develop a framework for gaze triggered events which canbe executed on a robot and mobile devices and allows to perform experiments.We experimentally evaluate the framework and techniques for extracting gazedirection based on a robot-mounted camera or a mobile-device camera whichare implemented in the framework. We investigate the impact of light on theaccuracy of gaze estimation, and also how the overall accuracy depends on usereye and head movements. Our research shows that the light intensity is im-portant, and the placement of light source is crucial. All the robot-mountedgaze detection modules we tested were found to be similar with regard to ac-curacy. The framework we developed was tested in a human-robot interactionexperiment involving a job-interview scenario. The flexible structure of thisscenario allowed us to test different components of the framework in variedreal-world scenarios, which was very useful for progressing towards our long-term research goal of designing intuitive gaze-based interfaces for human robotcommunication.