Online detection of driver distraction: preliminary results from the AIDE project

Abstract

A preliminary version of the Cockpit Activity Assessment (CAA) module, developed as a part of the AIDE EU-funded project, is described and evaluated. The CAA module is a real-time software implementing algorithms for online detection of visual and cognitive driver distraction. Algorithms for analyzing head/eye tracker output are presented, and are shown to be useful for visual distraction detection purposes although further developments are needed. The problem of cognitive distraction detection is addressed by suggesting three cognitive distraction indicators, defined so as to be robust to variations in sensor data quality, and shown to be individually sensitive to cognitive load in the driver. Finally, a support vector machine classifier, using the cognitive distraction indicators as input, is presented. On motorway, rural and suburban roads, the classifier currently reaches a 40-80 percent accuracy at detecting a cognitive task, while maintaining an almost 80 percent correct classification of non-distracted driving.