Eye-Tracking Using Image Processing Techniques and Deep Neural Network

Abstract

Eye-tracking technology measures pupil dilation, eye movements, and blinking, providing insights into visual attention. It monitors eyes in real time, converting data into gaze points, vectors, and pupil positions. The system includes cameras, lighting sources, and computational power. Advanced image processing algorithms process camera inputs into meaningful data. Eye-tracking has applications in detecting fatigue, biometric verification, assessing attention levels, and improving human-computer interaction. It is widely used in marketing, advertising, assistive technology, usability studies, clinical research, healthcare, education, and automotive sectors.

The study initially considered various methods, and previous research indicated that deep learning (DL) algorithms outperformed other methods. Therefore, a DL approach was chosen due to its ability to analyze and predict complex patterns in images, text, and sound. The study utilized a Dilated Eye Network constructed and trained on eye-tracking datasets using a deep neural network (DNN) architecture. To enhance accuracy, the pre-trained network was fine-tuned with the same datasets. Additionally, the Circular Hough Transform was employed to detect circular shapes, such as eyes identified by edge detection in images. The combined approach significantly improved outcomes, successfully detecting eyes in 17 out of 20 test images, showcasing positive results across the testing dataset.

This study presents an innovative eye-tracking system that combines a Dilated Eye Network with the Circular Hough Transform to detect eye movements. By leveraging DL to analyze complex visual patterns, this approach outperforms traditional methods that often yield inadequate results, particularly in varied conditions. The integration of the Circular Hough Transform enhances robustness by effectively identifying circular shapes in imperfect images. Overall, this research demonstrates superior performance and potential for diverse applications in marketing, healthcare, and human-computer interaction.

This study makes a unique contribution to eye segmentation and tracking by introducing the dilated eye network, which improves accuracy while preserving image resolution and minimizing computational load. Additionally, the integration of semantic segmentation and the Circular Hough Transform enhances performance in noisy and occluded environments. The significance of this research lies in its potential applications for effective eye health monitoring in younger populations, demonstrating a practical advancement that leverages innovative techniques alongside pre-trained networks to address real-world challenges.