A Comparative Exploration of Eye Gaze and Head Motion Cues for Lane Change Intent Prediction Anup Doshi and Mohan Trivedi Abstract— Driver behavioral cues may present a rich source of information and feedback for future Intelligent Driver Assis- tance Systems (IDAS). Two of the most useful cues might be eye gaze and head motion. Eye gaze provides a more accurate proxy than head motion for determining driver attention, whereas the measurement of head motion head motion as a derivative of pose is less cumbersome and more reliable in harsh driving con- ditions. With the design of a simple and robust IDAS in mind, we are interested in determining the most important driver cues for distinguishing Driver Intent. We use a lane change intent prediction system [1] to determine the relative usefulness of each cue for determining intent. Various combinations of input data are presented to a discriminative classifier, which is trained to output a prediction of probable lane change maneuver at a particular point in the future. Quantitative results using real- world data are presented and show that head motion, when combined with lane position and vehicle dynamics, is a reliable cue for lane change intent prediction. The addition of eye gaze does not improve performance as much as simpler head pose- based cues. I. I NTRODUCTION Intelligent driver assistance systems have the potential to save many lives by aiding drivers to make prompt, safe deci- sions about driving maneuvers. This year in the U.S. alone, over 43,000 fatalities are projected due to traffic collisions, with up to 80% of those due to driver inattention [2], [3]. To counter the effect of inattention, IDAS’s could be designed to provide the driver ample warning time to impending dangerous situations, and even assist the driver in reacting appropriately. The IDAS could thus prevent collisions and make roads safer. The basis of state-of-the-art IDAS systems today involve sensors detecting the environment outside the vehicle, along with the vehicle dynamics. Recent research has supported the incorporation of sensors looking inside the vehicle into these systems [4], [5]. A major advantage of monitoring drivers is the ability to observe driver behavior and potentially infer driver intent. We are interested in determining the important driver cues for distinguishing intent, in order to support future IDAS designs. In prior intent prediction research [1], [6], [7], head motion has been proposed as a pertinent cue. While robust monocular in-vehicle head pose estimation systems have been developed [8]–[10], it may be argued that head motion, as a derivative of pose, is not a sufficient estimate of true gaze. In order to derive precise gaze estimates, it follows that eye gaze should be included [11]. Unfortunately Authors with LISA: Laboratory for Safe and Intelligent Automobiles, University of California, San Diego, http://cvrr.ucsd.edu/lisa. {andoshi,mtrivedi}@ucsd.edu there are several drawbacks with modern eye-gaze estima- tors in vehicles, including the need to overcome lighting changes, shadows, occlusions, and potentially cumbersome stereo rigs or intrusive head-mounted cameras. Therefore we are motivated to determine if eye gaze and head motion are useful intent predictors, and furthermore which one (or combination) is the more informative cue. In this experiment, we use a lane change intent prediction system [1] to determine the relative usefulness of eye gaze and head motion data. Our comparative experiment is de- signed to distinguish the merits of the two cues and compare their importance. By determining the better cue, we hope to provide the basis for appropriate future designs of lane change intent systems, as well as a foundation for interactive driver assistance systems in general. II. DRIVER BEHAVIORAL CUES The analysis of driver behavior has long been a popular field of research in light of the potential for safety im- provements. NHTSA has most recently conducted studies of Driver Workload Metrics [3], including eye gaze as a proxy for driver workload. With respect to the particular maneuver of lane changes, the analysis of driver behaviors dates back at least 30 years. Here we present a summary of relevant research. We then present our methodology for determining driver behavior, in preparation for our comparative experi- ments. A. Lane Changes and Driver Visual Search According to early research in the field, there is significant reason to believe that behavior analysis of the driver can lead to reliable predictions about lane change intent. The time period three seconds ahead of the actual lane change was determined to be a critical time period during which the driver engages in a visual search to determine feasibility of lane change [12]. In fact according to Tijerina et al. [13], there are specific eye glance patterns which take place in the period before a lane change. It was determined that during left lane changes, there were between 65-85% chance of looking at the left mirror, and 56-67% chance of looking at the rearview mirror. Correspondingly, during right lane changes drivers looked at the right mirror with 36-53% probability, and the rearview mirror with 82-92% probability. Moreover the mirror glance duration before lane change manuevers lasted on average 1.1 seconds, varying between .8 and 1.6 seconds [2]. Mourant and Donohue observed that lengthy blind spot checks occured only in conjunction with 2008 IEEE Intelligent Vehicles Symposium Eindhoven University of Technology Eindhoven, The Netherlands, June 4-6, 2008 978-1-4244-2569-3/08/$20.00 ©2008 IEEE. 49 Authorized licensed use limited to: Univ of Calif San Diego. Downloaded on January 21, 2009 at 14:50 from IEEE Xplore. Restrictions apply.