Efficient Correlation Algorithm for Gaze Direction & Head Gesture

Abstract

The aim of this thesis is to explore new applications in the area of human computer interaction and to propose solution for these applications based upon gaze direction and head gesture. Gaze direction and head gesture are considered as input modalities for human computer interaction with different degree of freedom and different capabilities. Gaze direction estimation is achieved by subsequent stages: face detection, eye detection, eye gaze estimation and coordinate mapping for interaction of gaze over natural world surface. Face detection has been achieved by adaboost which combine visual critical feature based weak learner and produce a strong classifier. Assumingly face is detected, and then eyes are detected based upon texture feature. A regression neural network based gaze interaction with a surface is proposed. The regression neural network is trained over eye image while gazing in several directions. Accuracy of the proposed system is based upon the performance of this regression neural network that has to produce the coordinate which are being gazed by human eye. The detected eye gaze is further correlated with head gesture: head shake and head node to provide interaction mechanism with the real world. Practical performance of the system was tested in different real world environment such as infotainment device control and in automotive. The dissertation also proposes two novel applications in the area of augmented reality based upon gaze direction and head gesture. Augmented reality is combination of real viworld and computer generated data. A subset of gaze direction is proposed in which head orientation is considered and gross level gaze direction is proposed. This gross level gaze defines current field of view which is then animated and useful information is displayed for situation aware environments.