Abstract
The need for personal calibration has been one of the most significant technical limitations in video-based eye tracking, and there have been many research attempts on calibration-free techniques. This paper reviews calibration-free eye tracking techniques from two perspectives: person-independent gaze estimation and implicit personal calibration. Person-independent gaze estimation is the technique to infer gaze directions without considering person-dependent parameters. The methods in this category include both model-based approaches to estimate the optical axis angle and appearance-based approaches to obtain generic gaze estimation functions. Implicit personal calibration refers to the technique to adjust person-dependent parameters for better estimation accuracy without relying on any explicit calibration instruction. The methods in this category utilize various cues such as bottom-up fixation prediction, environmental/anatomical constraints, and correlation in temporal patterns to predict gaze target locations. This review summarizes representative approaches in both categories and discusses future perspectives and potential application scenarios.
