Online Classification of Eye Tracking Data for Automated Analysis of Traffic Hazard Perception

Complex and hazardous driving situations often arise with the delayed perception of traffic objects. To automatically detect whether such objects have been perceived by the driver, there is a need for techniques that can reliably recognize whether the driver's eyes have fixated or are pursuing the hazardous object (i.e., detecting fixations, saccades, and smooth pursuits from raw eye tracking data). This paper presents a system for analyzing the driver's visual behavior based on an adaptive online algorithm for detecting and distinguishing between fixation clusters, saccades, and smooth pursuits.