Uma abordagem para estimação remota do olhar sob luz visível em tempo real

Abstract

This thesis proposes an approach for the remote gaze estimation problem using a single camera. A solution for the head pose estimation problem is also suggested, indicating the possible gaze location. It has applicability in Human Computer Interaction, in monitoring driver drowsiness for avoiding road accidents, and in Content Image Based Retrieval field. Many applications have motivated the increasing investigation on gaze tracking on the past few years. Several approaches were proposed in the literature, many of those requiring specific hardware devices for working properly. The solution proposed in this thesis is based on the analysis of images acquired by ordinary camerasin visible lighting. In other words, no specific hardware device or even illumination constraints are imposed. The methodology was implemented and evaluated by real experiments with different individuals. Results coming from those experiments demonstrated thatthe system works properly in real time, obtaining the eyes location accurately. Our system was also applied in the remote control of a robotic head. We built heatmaps of the regions focused by users in the screen.