Calibration-Free Gaze Tracking Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional gaze tracking systems require calibration procedures and repeated recalibration, which can be inconvenient and time-consuming, and they often rely on glint information, limiting their accuracy and user experience.
Innovation Solution
A method and system for gaze tracking that uses a combination of machine learning and image-based techniques to track eye movements without the need for calibration, utilizing RGB-D video data from sensors like RGB-D cameras to accurately monitor gaze characteristics such as fixation, saccade, and pupil diameter, without relying on glint information, and can be implemented on mobile devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional gaze tracking systems use light sources to illuminate eyes and glint information to estimate gaze, then gaze tracking can be implemented, but calibration procedures are required which are time-consuming and inconvenient
Solution Approach 1:
The system performs self-calibration by automatically adapting to each user's eye characteristics through machine learning algorithms that process video data from standard cameras. The calibration process occurs autonomously without requiring manual intervention or predefined calibration points, enabling the system to serve itself and eliminate the need for time-consuming manual calibration procedures
Solution Approach 2:
The system changes the approach from using glint information (reflective parameters) to using direct video imagery parameters such as pupil center position, pupil radius, and eye geometry extracted through machine learning. This parameter transformation allows the system to achieve accurate gaze estimation without requiring traditional calibration to predefined spatial points
2Measurement precision
If conventional gaze tracking systems require repeated calibration each time gaze tracking is utilized, then accuracy can be maintained, but user convenience deteriorates
Solution Approach 1:
The system performs preliminary adaptation during initial use and continuously refines its gaze estimation models through ongoing machine learning processes. By preparing the calibration data structures and machine learning models in advance and updating them progressively, the system maintains high accuracy without requiring repeated interruption for recalibration
Solution Approach 2:
The system autonomously maintains and updates its calibration data through continuous processing of video input and machine learning adaptation. This self-maintaining capability ensures sustained measurement precision without requiring manual user intervention for repeated calibration procedures
3Measurement precision
If conventional gaze tracking systems use glint information for gaze estimation, then gaze direction can be determined, but system complexity increases due to light source requirements
Solution Approach 1:
The system extracts and removes the requirement for external light sources and glint detection from the gaze tracking system. By using standard video cameras to capture eye images and applying machine learning to directly estimate gaze parameters from these images, the system eliminates the complex illumination subsystem while maintaining or improving measurement precision
Solution Approach 2:
The system replaces the mechanical/optical illumination system (light sources and glint detection) with a computational approach using machine learning algorithms that process standard video imagery. This substitution eliminates the need for physical light sources and complex optical setups, reducing device complexity while achieving accurate gaze direction estimation
Data Source
AI summary
The present disclosure provides systems and methods for gaze tracking. The methods for gaze tracking may comprise (a) collecting video data of a subject's face using a device, and (b) processing the video data to track the subject's gaze. The video data may comprise a plurality of images containing depth information. The methods for gaze tracking may track the gaze of the subject without requiring any prior or subsequent calibration of the subject's gaze to a predefined calibration point in space having a known location. The systems for gaze tracking may be configured to implement the methods for gaze tracking.