Using Naturalistic Driving Performance Data to Develop an Empirically Defined Model of Distracted Driving

016533662016This research was sponsored by the U.S. Department of Transportation, University Transportation Centers Program.PDFTech ReportNEXTRANS Project No. 126UMUY2.1DTRT12-G-UTC05AlgorithmsAutomatic data collection systemsCellular telephonesData analysisDistractionResearch HubDistracted driversHidden Markov modelsVehicle kinematic dataNorth AmericaUnited StatesNextransUnited States. Dept. of Transportation. Research and Innovative Technology AdministrationBingham, C. RaymondBao, ShanFlannagan, CarolPradhan, Anuj K.University of Michigan. Transportation Research InstituteNTL-HIGHWAY/ROAD TRANSPORTATION-HIGHWAY/ROAD TRANSPORTATIONNTL-SAFETY AND SECURITY-SAFETY AND SECURITYNTL-SAFETY AND SECURITY-Human FactorsUS Transportation CollectionDriver distraction is defined as a diversion of attention away from the primary driving activity toward non-driving related tasks (Lee et al., 2008). Multiple resource theory (MRT) describes this diversion as a process of competition for attentional resources (Wickens, 2002). When the non-driving related tasks compete for the same resource (e.g., visual or cognitive), performance of the primary task is very likely to degrade. In 2009, the National Highway Traffic Safety Administration (NHTSA) reported highlights of analyses of crash databases for that year as related to distracted driving (Ascone, 2009). For example, in 2009, 5474 people were killed on U.S. roadways in motor vehicle crashes that were reported to have involved distracted driving. Of these, 18% (995) involved reports of cell phone as a distraction. Thus, cell phones were involved in approximately 3% of all fatalities. Of those injured in crashes in 2009, 20% involved reports of distraction. Of those, 5% involved cell phones. Thus, approximately 1% of injuries were reported as involving cell phones. Cell phone use and other driver distractions have been the subject of many studies resulting in a range of findings (Bao, Flannagan, & Sayer, submitted; Liang & Lee, 2010; Nemme & White, 2010; Redelmeier & Tibshirani, 1997; Strayer & Drews, 2007; Strayer & Johnston, 2001). However, the most challenging element of the science of driver distraction is that while most simulator studies clearly show performance deficits with secondary tasks (Drews, Yazdani, Godfrey, Cooper, & Strayer, 2009; Liang & Lee, 2010; Owens, McLaughlin, & Sudweeks, 2011), the crash data show steady decreases in total crashes, fatalities, and crash rates (IIHS, 2010; Ascone, 2009). One of the difficulties in understanding the effect of distraction, particularly cell phone use, on crashes has been that police reports have historically under-represented distraction or not coded various sources of distraction. As this issue has become more public, coding of distraction has increased in quantity and quality. The National Motor Vehicle Crash Causation Survey (NMVCCS) was conducted between 2005 and 2007 and involved in-depth investigation of the causation of a set of 6,949 crashes. At that time, 22% of drivers were distracted by one or more sources. Of these, 16% were conversing with a passenger and about 3.4% were either talking on or dialing a cell phone. Because in-depth investigations were done on-scene, these estimates are much less likely to be undercounting distraction. The objective of this study is to apply a stochastic modeling method, Hidden Markov Modeling method, to naturalistic driving data analysis, and to develop algorithms to identify distracted driving by using vehicle kinematic variables only