Machine Learning-Based Gaze Tracking Without Infrared Hardware
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gaze tracking technologies require additional equipment like infrared cameras and high magnification lenses, causing inconvenience and high costs, and suffer from processing speed degradation due to complex calculations.
Innovation Solution
A machine learning-based gaze tracking device that uses a camera module without a separate lighting device, employing a regression model to infer gaze coordinates based on detected feature points, including a face direction, eye-ball direction, and iris center, to track gaze without additional hardware.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional gaze tracking technology uses infrared cameras and high magnification lenses, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent extracts and removes the separate lighting device (infrared illuminator) from the traditional gaze tracking system. The invention uses only the camera module to capture images and perform gaze tracking by detecting feature points (pupil center, iris center, face direction) directly from ambient light images, eliminating the need for additional infrared lighting hardware while maintaining tracking functionality
Solution Approach 2:
The camera module is made multi-functional by enabling it to perform both image capture and gaze tracking without requiring separate specialized components. The same camera that captures facial images is used to detect eye features and calculate gaze direction through machine learning algorithms, making the single device serve multiple purposes
2Measurement precision
If traditional gaze tracking technology uses complex calculation methods, then measurement precision is improved, but processing speed deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models (regression models, neural networks) offline before actual gaze tracking operation. The models are trained using labeled datasets containing facial images and corresponding gaze coordinates, so that during real-time operation, the pre-trained models can quickly infer gaze positions without performing complex calculations, thus improving processing speed while maintaining precision
Solution Approach 2:
The patent replaces traditional mechanical/optical calculation methods with machine learning-based computational approaches. Instead of using complex geometric calculations and transformation matrices to determine gaze direction, the system uses trained neural networks and regression models that process images and output gaze coordinates through learned patterns, significantly reducing computational complexity and improving processing speed
Data Source
AI summary
A device and a method for gaze tracking based on machine learning are proposed. In one aspect, the device includes an input unit configured to input an image including a face, and a feature point detection processor configured to detect first feature points in the image including the face. The device may also include a face direction detection processor configured to detect a direction of the face based on the detected first feature points, and an eye-ball direction detection processor configured to detect an eye-ball direction, which is a feature of an eye-ball, from the detected first feature points. The device may further include a model training processor configured to train a gaze tracking model by using the detected first feature points and the eye-ball direction, and a gaze tracking processor configured to perform gaze tracking by using a trained gaze tracking model.