Person identification through entropy oriented mean shift clustering of human gaze patterns

The paper describes a system aimed at improving the human machine interaction that is able to identify users according how she looks at the monitor. The proposed system does not need invasive measurements that could limit the naturalness of her actions. The approach, here described, detects the gaze movements on the monitor and clusters the sequences of user gaze fixation points on the screen characterizing the user according the particular patterns her gaze follows. The recognition of the user is performed through a clustering process employing the Mean-Shift algorithm and it can open new perspective in human-machine interaction. In particular, the parameters of the clustering process are tuned optimizing an entropy oriented cost function that allows an automatic selection of the best parameters setting.