Drowsy Driver Detection Using Face Tracking Algorithms PDF Download

This study involved the collection of real-world driving data from a small sample of drivers, thought to be at heightened risk, to identify periods of drowsiness and inattention. Data included a variety of engineering measures including video of the driver and the road scene. One objective of the study was the identification of periods of drowsiness and inattention, documented on video, that would be made available for public education and outreach programs. A second objective was to validate, in a naturalistic driving setting, the drowsy driver detection algorithms developed by Wierwille, et al. in a simulator environment. Participants' personal vehicles were instrumented with the MicroDAS instrumentation system and all driving during the data collection was fully discretionary and independent of study objectives. The study thus offered the opportunity to implement highly unobtrusive data collection in subjects own vehicles with the absence of an experimenter in an effort to gather naturalistic data with a minimum of experimental artifacts. Results highlight the importance of lanekeeping variation as a key predictor variable for detecting drowsiness while driving, although the drowsy detection algorithm did not perform as well as in the simulator studies. An attempt to relate algorithm results to the prediction of driver inattention was inconclusive. The results are discussed in terms of theoretical, and procedural issues associated with inattention, drowsiness and driver responses to false positive epochs. It is suggested that the use of a multiplicity of approaches for addressing drowsy and inattentive driving would be most effective, and recommendations are made for future research on both technological and behavioral interventions