A wearable gaze tracking system for children in unconstrained environments

• Published in: Computer vision and image understanding Vol. 115; no. 4; pp. 476 - 486
• Main Authors: Noris, Basilio, Keller, Jean-Baptiste, Billard, Aude
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier Inc 01.04.2011; Elsevier
• Subjects: Adults; Appearance based; Applied sciences; Artificial intelligence; Calibration; Children; Computer science; control theory; systems; Computer systems and distributed systems. User interface; Data processing. List processing. Character string processing; Exact sciences and technology; Eye blinking; Gaze tracking; Illumination; Infants; Lay users; Lighting; Memory organisation. Data processing; Pattern recognition. Digital image processing. Computational geometry; Software; Support Vector Regression; Tracking errors; Tracking systems; Unconstrained environments; Unconstrained environments; Gaze tracking; Eye blinking; Support Vector Regression; Lay users; Appearance based; Children; Gaze; CCD camera; Lighting; Vector support machine; Illumination; Tracking error; Robustness; Child; Age; Human; Head; Target tracking; Peripheral vision; Regression analysis; Calibration; Pursuit eye movement; Luminance; Animal; Mirror
• DOI: 10.1016/j.cviu.2010.11.013
• ISSN: 1077-3142

► We perform eye-tracking from a head-mounted device designed for children. ► We use Support Vector Regression to map eye-image pixels to gaze coordinates. ► Passive lighting allows tracking on very wide field of view (96° × 96°). ► Tracking accuracy is higher for adults (1.59°) than for children (2.42°). We present here a head-mounted gaze tracking system for the study of visual behavior in unconstrained environments. The system is designed both for adults and for infants as young as 1 year of age. The system uses two CCD cameras to record a very wide field of view (96° × 96°) that allows to study both central and peripheral vision. A small motor-driven mirror allows to obtain the direction of the wearer’s gaze with no need for active lighting and with little intrusiveness. The calibration of the system is done offline allowing experiments to be conducted with subjects who cannot cooperate in a calibration phase (e.g. very young children, animals). We use illumination normalization to increase the robustness of the system, and eye blinking detection to avoid tracking errors. We use Support Vector Regression to estimate a mapping between the appearance of the eyes and the corresponding gaze direction. The system can be used successfully indoors as well as outdoors and reaches an accuracy of 1.59° with adults and 2.42° with children.