HexHeAd: 6D Head Pose Estimation Based Visual Focus of Attention Detection

This paper proposes a visual focus of attention detection method based on the user’s 6D head pose estimation, which determines the content of the observer’s visual focus of attention according to the current visual attention focus range of the observer. The system detects the observer’s current head posture and uses continuous tracking of the direction of interest to determine when the observer’s visual focus of attention is drawn to a specific location and content. Context-based visual focus of attention span is used as an approach to detect the level of interest an observer has in a particular content. Then, according to the level of interest of the observer, a visual focus of attention-based interaction channel can be established, and the observer’s feedback on specific content can be accumulated to obtain the relevant preferences of the observer and establish a foundation for future interactions. In order to evaluate the method, we conducted an experiment using short video as the target visual focus of attention, and when the observer watched a large screen playing specific content in an open area, the proposed method determined the current visual focus of attention level of the observer, and further discovered the interest of the observer according to the content viewed by the observer. The evaluation results show that our approach has good performance for automatic observer tracking or human-robot interaction.