Inferring Driver Intent: A Case Study in Lane-Change Detection

Abstract

This paper introduces a robust, real-time system for detecting driver lane changes. Under the framework of a “mind-tracking architecture,” the system simulates a set of possible driver intentions and their resulting behaviors using an approximation of a rigorous and validated model of driver behavior. The system compares these simulations with a driver's actual observed behavior, thus inferring the driver's unobservable intentions. The paper demonstrates how this system can detect a driver's intention to change lanes, achieving an accuracy of 85% with a false alarm rate of 4%; detecting 80% of lane changes within 1/2 second and 90% within 1 second; and detecting 90% before the vehicle moves 1/4 of the lane width laterally — that is, approximately when the vehicle first touches the destination lane line.