Advancements in driver distraction and driving performance assessment for robust in-vehicle systems

Driving a vehicle is a skillful and complicated task, requiring the driver to pay attention to the ever changing surrounding environment. However, many individuals perceive driving to be an extension of their natural skills, where the task is taken for granted. This complacency encourages drivers to multi-task while driving. With the proliferation of smart portable devices and other infotainment systems, along with the increased time spent in cars, multitasking has become common while driving. Though multitasking is probably inevitable, this competition for human resources introduces a variety of distractions that divert the driver's attention from the primary driving task leading to accidents. Though the automotive industry has made significant advancements in active safety systems, human error remains one of the major causes of accidents, resulting in enormous socio-economic losses. The current generation active safety systems utilize vehicle dynamics and environmental information, but are unaware of context and driver status and are unable to adapt quickly to changing situations. This dissertation addresses this problem by proposing a driver adaptive and context aware distraction detection system. The advancements are focused on identifying individual drivers, detecting variations in driving performance and providing feedback on driver state/distraction. The advanced system also attempts at isolating the source of distraction which adversely influences driving performance. A systematic analysis on the influence of secondary tasks shows that individual driver's comfort level plays a significant role. This entire system is evaluated using naturalistic driving UTDrive corpora. The dissertation also introduces the use of smart portable devices as an alternate form of instrumenting the vehicle. A careful utilization and delivery of sensor information from smart portable devices is shown to be more useful to the driver, contrary to the popular notion that bringing in smart portable devices into the car distracts drivers. The proposed system is also evaluated on portable device sensors showing that these devices could be effectively utilized in advanced driver assistance systems. The advancements stemming from this dissertation establish a new system approach to driver distraction analysis and modeling, which is most appropriate for current and future research in naturalistic driving for improved safety.