Date of Award

Winter 2012

Project Type

Dissertation

Program or Major

Electrical and Computer Engineering

Degree Name

Doctor of Philosophy

First Advisor

Andrew L Kun

Abstract

Driving is a cognitively loading task which requires drivers' full attention and coordination of both mind and body. However, drivers often engage in side activities which can negatively impact safety. A typical approach for analyzing the influences of side activities on driving is to conduct experiments in which various driving performance measures are collected, such as steering wheel angle and lane position. Those measures are then transformed, typically using means and variances, before being analyzed statistically. However, the problem is that those transformations perform averaging of the acquired data, which can result in missing short, but important events (such as glances directed off-road). As a consequence, statistically significant differences may not be observed between the tested conditions. Nevertheless, just because the influences of in-vehicle interactions do not show in the averages, it does not mean that they do not exist or should be neglected, especially if the nature of the interactions is such that they can be performed frequently (for example, with an infotainment system). This can create a false conclusion about the lack of influence of the tested side activity on driving.

The main contribution of this research is in developing two new performance measures inspired by the mathematical function of cross-correlation: one which evaluates the cumulative effect and the other which evaluates the effects of individual instances of in-vehicle interactions on driving and cognitive load. The results from three driving simulator studies demonstrate that our cumulative measure provides more sensitivity to the effects of in-vehicle interactions, even when they are not detected through average-based measures. Additionally, our instance-based measure provides a low-level insight into the nature of the influence of individual in-vehicle interactions. Both measures produce results that can be ranked, which allows determining the relative size of the effect that various in-vehicle interactions have on driving. Finally, we demonstrate a set of variables which can be used for predicting the cumulative and instance-based results. This predictive ability is important, because it may allow obtaining quick simulation results without performing actual experiments, which can be used in the early stages of an interface or experiment design process.

Download

Share

COinS