Brightness-Normalized CNNs for Eye Gaze Direction Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gaze estimation systems face challenges in accurately detecting pupil centers due to the non-rigid nature of the eye, frequent blinking, and iris color variations, especially when infrared cameras are not used, leading to gaze direction errors.
Innovation Solution
A computer-implemented method using a Convolutional Neural Network (CNN) model that adjusts image brightness and processes images through multiple layers to determine eye gaze direction, incorporating brightness correction and training techniques to minimize errors, and provides indications of eye gaze direction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional gaze estimation systems are used, then device complexity is reduced, but measurement precision deteriorates due to inability to accurately detect pupil center
Solution Approach 1:
The patent replaces traditional mechanical/optical gaze estimation systems with a CNN-based computational model. The system uses a Convolutional Neural Network to directly process images and predict gaze direction, eliminating the need for complex pupil center detection algorithms and infrared cameras. This substitution of mechanical systems with an AI-based computational approach achieves high measurement precision while reducing overall device complexity.
Solution Approach 2:
The patent transforms the input image parameters by adjusting brightness and contrast to enhance pupil visibility. The system applies image processing techniques to normalize illumination conditions and highlight the pupil region, making it more detectable for the CNN model. This parameter transformation enables accurate gaze detection without requiring complex hardware modifications.
2Measurement precision
If infrared cameras are used to capture eye images, then measurement precision improves for pupil detection, but device complexity increases
Solution Approach 1:
The patent replaces infrared cameras with standard visible light cameras combined with CNN-based processing. Instead of using infrared radiation to enhance pupil visibility, the system uses a Convolutional Neural Network to learn features from visible light images, automatically identifying pupil centers and calculating gaze direction. This substitution eliminates the need for specialized infrared hardware while maintaining or improving detection accuracy.
Solution Approach 2:
The patent creates a computational model that copies and replicates the functionality of infrared cameras through software-based image processing. The CNN model learns to extract pupil information from standard camera images by processing visual features, effectively copying the pupil-enhancing function of infrared cameras without requiring the physical hardware. This allows the system to achieve similar or better performance using more accessible imaging devices.
3Measurement precision
If image brightness is not normalized, then ease of operation is maintained, but measurement precision deteriorates due to lighting variations
Solution Approach 1:
The patent applies preliminary image processing steps before feeding images to the CNN model. The system performs brightness normalization and contrast enhancement in advance to ensure consistent input conditions. By preparing and pre-processing images beforehand, the system eliminates the need for complex real-time adjustments during operation, maintaining ease of use while improving measurement precision through standardized input conditions.
Data Source
AI summary
Methods and systems for detecting eye gaze direction in digital images. Digital images can be brightness corrected prior to being processed by a convoluted neural network trained for eye gaze detection. Brightness correction can result in greater accuracy by providing images of similar brightness to the images the convoluted neural network was trained on. The convolutional neural network can use multiple convolution layers, fully connected layers, batch normalization, rectified linear activation functions, and dropout to process an image for gaze determination.