What are you looking at? Automatic estimation and inference of gaze

This thesis is focused on gaze estimation, that is, providing an automated answer to the question What are you looking at?. To this end, the problem of gaze estimation is divided into three sub problems, namely detection (i.e. extracting the relevant visual cues from the faces), estimation (i.e. combining them to compute a rough line of interest), and inference (i.e. using information about the gazed scene to infer the most probable gazed target). With the final goal of achieving a more accurate and natural gaze estimation system, this thesis focuses on improving eye center location, head pose estimation and image saliency methods, and combining them into a unified framework.