An eye movement recognition algorithm for ocular occlusion scenarios

Eye movement recognition technology has made remarkable progress in both entertainment and medical field. Presently, two prevalent approaches are in use: Electrooculography (EOG)-based and image-based eye movement recognition. Nonetheless, each method comes with its own set of challenges. EOG-based recognition needs specialized equipment for data collection. In addition, traditional image-based eye movement recognition algorithms cannot work when the eyes are occluded. To tackle these obstacles, we propose an improved image-based eye movement recognition algorithm that can achieve eye movement recognition even when the eyes are occluded. We enhance the subtle changes through Eulerian amplification, convert the periocular images into one-dimensional waveforms, and use the peak detection algorithm to obtain eye movement signals from the output of independent component analysis (ICA). Finally, a network combined CNN and Bi-LSTM is proposed to recognize the eye movement. The experimental results show that the recognition rate of the proposed method can reach 72.46% even when the eyes are occluded.