Integrating the saliency map with distract-r to assess driver distraction of vehicle displays.

摘要

There are a growing number of potential distractions in vehicles today, such as navigation, collision warning, and entertainment systems. These systems promise substantial benefits for driving comfort, efficiency and safety, but they might also distract drivers. This dissertation develops computational cognitive models of driver behavior to assess the distraction potential of vehicle displays. One of the main goals of this dissertation is to integrate a saliency-based model, a saliency map, into Distract-R to build a computational model that can account for top-down and bottom-up attentional influences. The saliency-based model quantifies exogenous influences (e.g., visual features of a display) of visual attention while Distract-R quantifies endogenous influences (e.g., drivers' goals and expectations) of visual attention with respect to secondary tasks and vehicle displays. Two experiments were conducted to guide model development and to validate model predications. The experiments showed that design features of vehicle displays affected driving performance and glance duration to the secondary task, and both top-down and bottom-up attentional processes were engaged when drivers interacted with driver-vehicle interfaces. To integrate Distract-R and the saliency map, activation fields that describe the interaction between top-down and bottom-up attentional process were used to determine glance duration to the display. The integrated model was validated with empirical data, showing that the model could predict drivers' pattern of glance durations to a level comparable to between-subject variability--the theoretical limit of prediction. This dissertation contributes to modeling driver distraction by integrating two models to account both top-down and bottom-up influence on visual attention, and by building a tool for assessing the potential distraction of vehicle displays.