Automatic social attention analysis in infants

Abstract

In recent years, eye-tracking technologies have been evolving rapidly due to their significant importance in multi-disciplinary fields. These advancements have eased the tracking of visual attention and invited a focus of research utilizing eye-tracking in the assessment of infant attention and cognitive development. However, the analysis processes of eye-tracking data are not up to date with the current technologies. The majority of the studies that examine infants’ attention to social scenes follow a manual or semi-manual approach to annotate the regions of interest (ROI) in the visual stimuli and perform analysis of eye-tracking data. The traditional approaches adopted by researches are time-consuming and impose limitations on research directions, for instance, limitations on ROI extraction from complex visual scenes stimuli. This has motivated the present work towards designing an automated pipeline for tackling the limitations of the traditional approaches. Our proposed system uses computer vision models to analyze eye-tracking data towards the faces in visual data, to generate ROIs, and to parameterize the gaze behavior of the measured subject. We test our pipeline and parameters on clinical eye-tracking data collected from infants at 7 and 16 months of age in rural South Africa. Results from the experiments demonstrate the feasibility of the developed automated pipeline in detecting ROIs and show that the extracted gaze parameters are sensitive to the age of the subjects—a proxy for cognitive development. Moreover, we have found consistent results related to previous research on the general gaze behavior of infants, in terms of observing a bias of visual attention towards faces. In addition, we tested the ability of our parameters to predict cognitive measures using a linear regression model. The obtained results were not of high accuracy. However, the proposed automated pipeline and parametrization can still be applied efficiently in eye-tracking research to conserve time and expand the area of research, e.g., to use of complex social scenes stimuli with faces.