Comparing Three Methods of Eye-Glance Transition Coding

In order for researchers to analyze driver glance behavior and compare their results to other studies, glance evaluation methodologies must be comparable in terms of how specific terms are defined. To this end, the International Organization for Standardization (ISO) developed the ISO 15007 standard, which was adopted by the Society of Automotive Engineers (SAE) as SAE J2396. This report describes an experiment designed to test the validity and flexibility of one key aspect of these standards with respect to how transitions between glances may be assessed. Three glance evaluation methodologies were tested: the Destination Method (DM), the Origin Method (OM), and the Transition Method (TM). Ninety epochs were selected from the Second Strategic Highway Research Program (SHRP 2) Naturalistic Driving Study data set. The epochs were divided into three sets of 30 epochs with similar complexity. The three sets were then assigned to one of the three eye-glance methods. Five trained eye-glance analysts and one expert rater coded the three sets of epochs using the assigned method, and the resulting glance data were then analyzed to compare the time required to code for each method, the inter- and expert-rater reliability of the three methods, and the effect of coding method on key glance metrics. The results show that the OM was the fastest manual eye-glance coding method (with a recommended time multiplier of 10:1), followed by the DM (12:1), and then the TM (14:1). All three methods produced similar inter-rater and expert-rater reliability scores, averaging about 88%. The three methods of coding also yielded similar key glance metrics, such as average percentage of time not forward, mean forward glance duration, and mean not forward glance duration.