Student Eye Gaze Tracking and Attention Analysis System using Computer Vision

Along with self-reports and periodic assessments students' engagement in online lectures can be understood with the help of eye gaze tracking. This data can provide valuable insights on the learning pattern of students and areas of difficulty which further aid the teachers in improving their teaching style. Extensive research has been performed in eye gaze tracking based on their location, shape, texture, or a combination of these features. Usually, deep learning-based neural networks are used for this purpose due to its increased efficiency and accuracy. Deep learning requires huge amount of high-quality annotated data to train the model and has high complexity. Thus, this paper proposes an approach for analyzing the attention of students in online classes by tracking the eye gaze using computer vision which is computationally cheap. This technique is being used in various applications such as human-computer interaction, cognitive psychology, and security systems. Using eye-tracking technology, researchers can follow and measure eye movements, pupil dilation, point of glance, and blinking to identify where study subjects' visual attention is focused, what they pay attention to, and what they ignore. With the help of Haar Cascade and pupil detection algorithms such as Daugman algorithm, Hough Circular Transform, Blob Detection, and Centroid Detection, the eye movements are tracked. In order to map the eye gaze onto the monitor's screen a geometric model is proposed which can classify the gaze direction into nine states with respect to the screen: center, left, right, top, top left, top right, bottom, bottom left, and bottom right. The system calculates the amount of time a student is not paying attention to the screen based on which the attentiveness in online lectures is decided. There are three classifications for students' attention states: alert, sleepy, and not attentive. This system is sensitive to glares, head movements, varying lightings, and motion blur due to the lower accuracy and difficulty in detecting the pupil by computer vision algorithms.