Mapping of Human Visual Attention from Head Pose Measurements

In a human-robot interaction scenario, the robot must be aware of which objects capture the user’s attention. From mapping and tracking the human attention, the robot can retrieve information such as the position of objects in the scene. Our approach can estimate the user’s areas of attention in the environment during the execution of task (e.g. working at the computer, reading a book). The environment model consist of a point cloud, while the user’s field of attention is retrieved by tracking of the head pose. The user’s head pose is estimated through camera, avoiding the use of external, intrusive devices such as gaze tracking glasses. However, the user’s capability to act in an unconstrained fashion degrades the quality of the measurements. In spite of that, our algorithm can robustly segments the surrounding environment according to the user’s attention and estimates the centres and volumes of each area