Crash/Near-Crash: Impact of Secondary Tasks and Real-Time Detection of Distracted Driving

The main goal of this dissertation is to investigate the problem of distracted driving from two different perspectives. First, the identification of possible sources of distraction and their associated crash/near-crash risk. That can assist government officials toward more informed decision-making process, allowing for optimized allocation of available resources to reduce roadway crashes and improve traffic safety. Second, actively counteracting the distracted driving phenomenon by quantitative evaluation of eye glance patterns. This dissertation research consists of two different parts. The first part provides an in-depth analysis for the increased crash/near-crash risk associated with different secondary task activities using the largest real-world naturalistic driving dataset (SHRP2 Naturalistic Driving Study). Several statistical and data mining techniques are developed to analyze the distracted driving and crash risk. More specifically, two different models were employed to quantify the increased risk associated with each secondary task: a baseline-category logit model, and a rule mining association model. The baseline-category logit model identified the increased risk in terms of odds ratios, while the A-priori association algorithm detected the associated risks in terms of rules. Each rule was then evaluated based on the lift index. The two models succeeded in ranking all the secondary task activities according to the associated increased crash/near-crash risk efficiently. To actively counteract to the distracted driving phenomenon, a new approach was developed to analyze eye glance patterns and quantify distracted driving behavior under safety and non-Safety Critical Events (SCEs). This approach is then applied to the Naturalistic Engagement in Secondary Tasks (NEST) dataset to investigate how drivers allocate their attention while driving, especially while distracted. The analysis revealed that distracted driving behavior can be well characterized using two new distraction risk indicators. Additional statistical analyses showed that the two indicators increase significantly for SCE compared to normal driving events. Consequently, an artificial neural network (ANN) model was developed to test the SCEs predictability power when accounting for the two new indicators. The ANN model was able to predict the SCEs with an overall accuracy of 96.1%. This outcome can help build reliable algorithms for in-vehicle driving assistance systems to alert drivers before SCEs