Online Proctoring using Eye Gaze Tracking and Head Pose Estimation

With the emergence of COVID-19, education systems have switched over to elearning. Online examinations are currently held by utilizing webcams and supervisors, but limitations arise with the growing number of students. Our system detects the examinee’s gaze direction and estimates the examinee’s head position in real-time for online proctoring. The eye region obtained using facial landmarks is used in the subsequent stages to classify the eye gaze direction. The eye gaze space is reduced to three classes: left, right and center. A CNN is employed in this work for the classification of eye gaze direction and achieved an accuracy of 97%. The head pose is estimated by projecting the camera coordinates onto the image plane. The head position along with the eye gaze direction is used to identify unfair practices of the examinee. The application saves the person's data when his VFOA (visual focus of attention) is off the screen for the examiner to manually check and mark whether the examinee's action was attempted malpractice or just a momentary lapse in concentration. Thus, our application successfully identifies suspicious activity and alerts the examinee.