Realtime AAM based user attention estimation

In this paper a method of automatic real-time capable visual user attention for a face to face human machine interaction is described. This method based on Active Appearance Models (AAMs) and Multilayer Perceptrons (MLPs) to map the Active Appearance Parameters (AAM-Parameters) onto the current head pose. Afterwards, the chronology of the head pose becomes classified to attention or inattention. This visual attention estimation will be used in service robotic by human-robotic interaction to get a feedback whether the user is interested in the current dialog and for correct interpretation of the current emotional condition. To allow a more natural dialog the head pose is also very efficient interpreted as head nodding or shaking by the use of adaptive statistical moments. Especially, the head movement of many demented people are restricted, so they often only use their eyes to look around. For that reason this paper examine a simple gaze estimation with the help of an ordinary webcam.