Evaluation of Strategies for Integrated Classification of Visual-Manual and Cognitive Distractions in Driving

Prior studies have demonstrated unique driver behavior outcomes when visual and cognitive distraction occurs simultaneously as compared to the occurrence of one form of distraction alone. This situation implies additional complexity for the design of robust distraction detection systems and vehicle automation for hazard mitigation.

This study evaluated the effectiveness of two distraction classification strategies: (a) a "two-stage" classifier, first detecting visual-manual distraction and then identifying dual or cognitive distraction states, and (b) a "direct-mapping" classifier developed to identify all distraction states at the same time.

Driving performance data were collected on 20 participants under different known states of distraction (none, visual-manual, cognitive, and combined). A support vector machine (SVM) was used as a base algorithm for both classifiers and performance data as well as the level of driving control (tactical and operational), which served as inputs and modifiers to the classification process.

The two-stage strategy was found to be sensitive for identifying states of visual-manual distraction; however, the strategy also produced a higher false alarm rate than direct-mapping. Consideration of driving control levels during classification also improved classification accuracy. Future work needs to account for strategic levels of vehicle control.