Distraction Detection Algorithm Evaluation. Traffic Safety Facts.

摘要

In the past 10 years, several algorithms for detecting distraction have emerged. However, there has been no uniform method for assessing and comparing these algorithms to identify which algorithms are most promising and what interventions each algorithm might support. This study demonstrates a protocol for distraction detection algorithm assessment. The protocol consists of a data collection process that samples a selection of drivers 25 to 50 years old, driving situations (urban, rural, freeway), and representative distractions (turning, looking and reaching, looking and touching, and cognitive) designed to challenge the algorithms in a variety of ways and reveal their capabilities and vulnerabilities. The data was collected using a high-fidelity, motion-based driving simulator (NADS-1) equipped with eye- and head-tracking hardware; active feedback on steering wheel, brake pedal, and accelerator pedal; and a fully operational dashboard. Data were interpreted relative to evaluation metrics from signal detection theory.