Eye-movement Analysis and Prediction using Deep Learning Techniques and Kalman Filter

Abstract

Eye movement analysis has gained significant attention from the eye-tracking research community, particularly for real-time applications. Eye movement prediction is predominantly required for the improvement of sensor lag. The previously introduced eye-movement approaches focused on classifying eye movements into two categories: saccades and non-saccades. Although these approaches are practical and relatively simple, they confuse fixations and smooth pursuit by putting them up within the non-saccadic category. Moreover, Eye movement analysis has been integrated into different applications, including psychology, neuroscience, human attention analysis, industrial engineering, marketing, advertising, etc. This paper introduces a low-cost eyemovement analysis system using Convolutional Neural Network (CCN) techniques and the Kalman filter to estimate and analyze eye position. The experiment results reveal that the proposed system can accurately classify and predict eye movements and detect pupil position in frames, notwithstanding the face tracking and detection. Additionally, the obtained results revealed that the overall performance of the proposed system is more efficient and effective comparing to Recurrent Neural Network (RNN).