Recognizing the Visual Focus of Attention for Human Robot Interaction

We address the recognition of people's visual focus of attention (VFOA), the discrete version of gaze that indicates who is looking at whom or what. As a good indicator of addressee-hood (who speaks to whom, and in particular is a person speaking to the robot) and of people's interest, VFOA is an important cue for supporting dialog modelling in Human-Robot interactions involving multiple persons. In absence of high definition images, we rely on people's head pose to recognize the VFOA. Rather than assuming a fixed mapping between head pose directions and gaze target directions, we investigate models that perform a dynamic (temporal) mapping implicitly accounting for varying body/shoulder orientations of a person over time, as well as unsupervised adaptation. Evaluated on a public dataset and on data recorded with the humanoid robot Nao, the method exhibit better adaptivity and versatility producing equal or better performance than a state-of-the-art approach, while the proposed unsupervised adaptation does not improve results.