Abstract

Gaze and head pose estimation can play essential roles in various applications, such as human attention recognition and behavior analysis. Most of the deep neural network-based gaze estimation techniques use supervised regression techniques where features are extracted from eye images by neural networks and regress 3D gaze vectors. I plan to apply the geometric features of the eyes to determine the gaze vectors of observers relying on the concepts of 3D multiple view geometry. We develop an end to-end CNN framework for gaze estimation using 3D geometric constraints under semi-supervised and unsupervised settings and compare the results. We explore the mathematics behind the concepts of Homography and Structure-from- Motion and extend it to the gaze estimation problem using the eye region landmarks. We demonstrate the necessity of the application of 3D eye region landmarks for implementing the 3D geometry-based algorithms and address the problem when lacking the depth parameters in the gaze estimation datasets. We further explore the use of Convolutional Neural Networks (CNNs) to develop an end-to-end learning-based framework, which takes in sequential eye images to estimate the relative gaze changes of observers. We use a depth network for performing monocular image depth estimation of the eye region landmarks, which are further utilized by the pose network to estimate the relative gaze change using view synthesis constraints of the iris regions. We further explore CNN frameworks to estimate the relative changes in homography matrices between sequential eye images based on the eye region landmarks to estimate the pose of the iris and hence determine the relative change in the gaze of the observer. We compare and analyze the results obtained from mathematical calculations and deep neural network-based methods. We further compare the performance of the proposed CNN scheme with the state-of-the-art regression-based methods for gaze estimation. Future work involves extending the end-to-end pipeline as an unsupervised framework for gaze estimation in the wild.

Library of Congress Subject Headings

Gaze--Data processing; Neural networks (Computer science); Convolutions (Mathematics); Eye tracking; Machine learning

Publication Date

7-2020

Document Type

Thesis

Student Type

Graduate

Degree Name

Imaging Science (Ph.D.)

Department, Program, or Center

Chester F. Carlson Center for Imaging Science (COS)

Advisor

Guoyu Lu

Advisor/Committee Member

Jeff Pelz

Advisor/Committee Member

Carl Salvaggio

Campus

RIT – Main Campus

Plan Codes

IMGS-PHD

Share

COinS