SESUG 2011 1 Paper ST-12 Eyes on the Road: A Methodology for Analyzing Complex Eye Tracking Data Mary Anne Bertola and Stacy A. Balk Science Applications International Corporation (SAIC), McLean, VA ABSTRACT Distracted driving is a relevant social issue with potentially devastating consequences. In part due to recent calls from President Obama and United States Transportation Secretary LaHood to curb distracted driving, research on the topic is becoming more prevalent. The use of eye tracking devices in on-road vehicles is an invaluable resource to investigate driver situational awareness and attention capture. Such tools provide insight into where drivers are looking, both within and outside the vehicle, while traveling down a roadway. Data from eye trackers in a real world environment, however, present a unique set of analysis challenges. For example, there are multiple ways to quantify visual behavior (e.g., duration of fixations, percentage of time, etc.) and such quantifications are constrained to non- negative values since a driver cannot look at an object for a negative amount of time. Additionally, responses are correlated since it is general practice to use eye movement data from one person over a period of time, as opposed to one specific instance in time. The GENMOD procedure in SAS ® lends itself to accommodating such analysis challenges of eye tracking data through the use of generalized estimating equations which allow for restrictions on the values of a response variable and account for correlated measurements. This paper demonstrates the application of generalized estimating equations through the GENMOD procedure to analyze driver visual behavior in the presence of different roadway environments. Eye tracking devices are implemented in a variety of settings (e.g., training flight simulators, software usability, etc.). As such, it is hoped that analytical methodologies presented in this paper are also useful in the analysis of a variety of other eye tracking applications. INTRODUCTION Eye tracking is a relatively simple methodology that has long been used as a way to quantify eye movements. Early eye tracking devices commonly involved participants wearing contacts with embedded wiring that covered a large portion of the sclera (the white of the eye). Not only were these devices cumbersome, they also had the potential to be quite dangerous. However, most modern equipment utilizes simple active infrared light (not visible to humans) to create reflections from the eye that are captured by small camera systems. This setup provides the ability to unobtrusively, and relatively inexpensively, track eye movements (see Duchowski, 2007 for overview). Advancements in technology have not only provided a simpler way to investigate eye movements, but have also provided mobility that has allowed eye trackers to be implemented in a wide variety of environments. One such example is the integration of an eye tracking system which allows the eyes (and often head movements) to act as computer mouse and keyboard controls (e.g., Chin, Barreto, Cremades, & Adjouadi, 2008). This integration provides the opportunity for those people who do not have functional use of their hands to use the computer in a more naturalistic way than might otherwise be possible. Eye trackers can also be integrated to create gaze-contingent fisheye effects on computer screens. That is, an area of a screen can be visually enhanced or enlarged by simply shifting visual attention (e.g., Ashmore, Duchowski, & Shoemaker, 2005). This technique might be helpful to those with severe visual impairments or those who must closely visually inspect images (e.g., x-rays). Beyond these, eye tracking can be used in a variety of marketing, human factors, and ergonomics applications. For example, eye tracking can tell researchers where consumers look on a webpage, how people scan for specific objects in a cluttered array, or even how pilots visually search cockpits. This visual scanning information has the potential to be used to enhance designs to best capture observer attention, when and where it is appropriate. While eye tracking has long been used in indoor or predictable viewing environments (Young & Sheena, 1975), implementation in more dynamic environments is relatively recent. One such environment is the on-road vehicle. The remainder of this paper focuses on eye tracking and eye tracking data analysis in the automobile. Further, it investigates a case study involving data from an instrumented vehicle. In-vehicle eye tracking systems have the potential to provide insight into where and how long drivers are looking at environmental objects, both inside and outside of the vehicle. For example, it has been shown that visual patterns differ between novice and experienced drivers (e.g., Mourant & Rockwell, 1972). Findings such as these can be applied to teaching novice drivers where visual attention should be directed while driving. Beyond this, in-vehicle eye trackers even have the ability to detect drowsy driving in real time (Barr, Howarth, Popkin, & Carroll, 2005). More pertinent to this paper, though, is the capability of eye trackers to perform real time calculation of drivers’ time looking away from the road, which could potentially indicate in-vehicle distracted driving (Ahlström & Kircher, 2010).