부산대학교 도서관

Machine Learning (ML) techniques, specifically Support Vector Machine (SVM) and Extreme Gradient Boosting (XGBoost), were employed to achieve precise and intuitive real-time eye tracking mouse control through computer vision. However, XGBoost may suffer from overfitting when dealing with a large number of features compared to the training data size, or when dealing with noisy or imbalanced data. To address this issue, this paper introduces a Recurrent Neural Network (RNN), for eye tracking mouse control in Human Computer Interaction (HCI). Gaze Direction Estimation (GDE) is initially employed to estimate gaze direction, utilizing pupil positions and camera calibration parameters. The estimated gaze direction is then used as input for the RNN to eye-tracking mouse control for HCI. The experimental results shows that the GDE-RNN has 26.03% and 8.24% superior accuracy, 24.93% and 86.76% better precision, and 27.36% and 8.75% high recall in comparison to SVM and XGBoost for eye tracking mouse control.