Webcam-based gaze estimation for computer screen interaction.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Machine Learning

Background:

• Traditional eye tracking solutions often require specialized hardware.
• Webcam-based gaze estimation faces challenges with head movements and accurate 3D positioning.
• Existing methods may lack accuracy or require additional equipment.

Purpose of the Study:

• To develop and evaluate a novel, markerless, webcam-based gaze estimation system.
• To accurately map user gaze vectors and determine 3D position on computer screens.
• To compare the proposed system's performance against high-end hardware and existing software solutions.

Main Methods:

• Appearance-based gaze estimation models were utilized.
• A 2D webcam was employed for data acquisition without extra markers.
• Structure from Motion (SfM) was integrated into a Convolutional Neural Network (CNN) model to compensate for head movements.

Main Results:

• The system demonstrated accurate screen gaze tracking using only a standard webcam.
• A novel method for computing physical distance from the screen was presented.
• The integration of SfM effectively compensated for head yaw rotations, a common challenge.
• Comparative analysis showed competitive performance against dedicated eye trackers and software.

Conclusions:

• Accurate gaze estimation and 3D positioning are achievable with a single, standard webcam.
• The proposed SfM-CNN integration offers a robust solution for head movement compensation in gaze tracking.
• This research paves the way for accessible and advanced real-world gaze estimation applications.