Measuring inferred gaze direction to support analysis of people in a meeting

This paper introduces a method to infer gaze direction and the point of focus of participants involved in a collaborative activity, such as a meeting. It uses a single depth-based sensor placed overhead to capture the meeting, which has the benefit of avoiding occlusion and is unobtrusive, minimising possible changes in behaviour that might arise if people are aware of the sensor. The inferred gaze direction of each participant is estimated in the horizontal plane from the orientation of the head (yaw), derived from a segmentation of the depth image to generate an outline of the head. A common focus of attention is inferred by intersecting the gaze directions of each participant. Performance evaluation using a depth camera to record a meeting achieved a head detection performance of 99.6% and a valid gaze detection of 96.9%.