Traditional video communication systems offer a very limited experience of eye contact due to the offset between cameras and the screen. In response, we present EyeGaze, which uses multiple Kinect cameras to generate a 3D model of the user, and then renders a virtual camera angle giving the user an experience of eye contact. As a novel approach, we use concepts from KinectFusion, such as a volumetric voxel data representation and GPU accelerated ray tracing for viewpoint rendering. This achieves detail from a noisy source, and allows the real-time video output to be a composite of old and new data. We frame our work in literature on eye contact and previous approaches to supporting it over video. We then describe EyeGaze, and an empirical study comparing it with communication face-to-face or over traditional video. The study shows that while face-to-face is still superior, EyeGaze has added value over traditional video in terms of eye contact, involvement, turn-taking and co-presence.
