Eye Contact Detectability Thresholds for Mixed Reality Telepresence

As hybrid telepresence become more pervasive, preserving eye contact between interlocutors becomes paramount for regulating social interactions. Prior work has explored eye contact dynamics in videoconferencing systems, yet eye contact detectability remains under-characterized in mixed reality (MR) telepresence. This work analyses how gaze-direction misalignments disrupt eye contact perception within a 2D-panel-based MR telepresence configuration, which can occur when systems require users to focus on where interlocutors are displayed rather than on the acquisition camera. We calculate eye contact detectability thresholds through a two-alternative forced choice (2AFC) user study (N = 30 per set), analyzing how physical camera offsets along the X and Y axes influence eye contact perception. Participants in an MR environment viewed pairs of videos of interlocutors and indicated which showed better eye contact, one with no camera offset, or another with a camera offset in the [-10 cm, 10 cm] interval. We measured detectability thresholds for interlocutors with and without glasses, with dark and light irises. Findings suggest that detectability varied more systematically with iris color than with glasses, with patterns differing between the horizontal and vertical axes. These insights provide guidelines for understanding eye contact dynamics in MR telepresence. We also report a preliminary exploration on geometric transformations-rotating (yaw) the virtual panel displaying the interlocutor-can compensate for gaze misalignments in MR. Preliminary results suggest that panel rotation may partially restore perceived eye contact for moderate camera-display offsets.