MutualGazeReader: Automatic mutual gaze detection in face-to-face dyadic interaction videos demonstration showcase
Organiser: Cristina Palmero, Noldus Information Technology & KU Leuven
Mutual gaze, also known as eye contact, is an important non-verbal cue in social signal processing and a promising measure of the quality of interaction, particularly playing a crucial role in parent-child interactions. Eye gaze behavior, as an indicator of human visual attention, is also analyzed to assess communication skills, such as in turn-taking or during group meetings, and to detect possible behavioral disorders such as autism and ADHD.
In this session, we will introduce MutualGazeReader, a prototype for automatic mutual gaze detection for face-to-face dyadic interaction videos. Assuming a calibrated 2-camera setup, the system first estimates the line of gaze for each of the interactants using a novel state-of-the-art deep learning approach, and then combines it, along with the interactants location in the 3D scene, to detect ‘Looking at other’s face’ events. The system creates an output file with the behavioral state data, which can be exported to event logging applications such as The Observer XT.