Object Selection Using LSTM Networks for Spontaneous Gaze-Based Interaction

Two years on with Covid-19, touchless technology has evolved from a device that symbolizes luxury to something that is necessary. Eye tracker is one type of touchless technologies that uses user's gaze to interact with computer without touching the screen. Development of spontaneous gazebased interaction is progressing very rapidly. Researchers have developed various object selection methods without prior gazeto-screen calibration. Recently, the conventional approach of setting threshold was developed as a gaze-based object selection method. However, the use of threshold values is considered non-adaptive and requires additional data pre-processing to handle noises. To overcome this problem, deep learning is used as an object selection method for spontaneous gaze-based interaction. Deep learning does not require any data preprocessing method to achieve accurate object selection results. Out of five deep learning algorithms that were evaluated, LSTM (Long Short-Term Memory) and BiLSTM (Bidirectional Long Short-Term Memory) networks achieved comparable accuracy of $95.17 \pm 0.95$% and $95.15 \pm 1.17$%, respectively. In future, our research is promising for development of real-time object selection technique for touchless public display.