Reducing Fixation Error Due to Natural Head Movement in a Webcam-Based Eye-Tracking Method

Manuela Kunz, Arsalan Syed, Kathleen C. Fraser, Bruce Wallace, Rafik Goubran, Frank Knoefel, Neil Thomas

Abstract: The detection and monitoring of cognitive decline has emerged as an important research topic as the global population ages. Here, we present work towards the goal of using a webcam as a low-cost and non-invasive sensor to detect oculomotor changes associated with prodromal or early dementia. Specifically, we implement a method of 3D gaze tracking to account for natural head movement during the data collection. In a user study, we show that this method decreases gaze estimation error in a fixation task by 11 %, when compared with a standard 2D method. Performance improvements are greater in the horizontal direction than vertical direction, indicating the predominance of horizontal head movements in our users. Correcting for naturalistic head movements will be critical in the deployment of this technology, particularly in older user populations.