Method for Identifying Vehicle Movements for Analysis of Field Operational Test Data

Abstract

The independent evaluation of a roadway-departure crash-warning system will involve identifying and classifying conflicts and near-collisions from a vast quantity of field data. The classification of these events includes identifying vehicle movements such as negotiating a curve, making a turn, changing lanes, and merging. A unified algorithm that relies solely on vehicle kinematic data is developed for identifying such movements. The method works reliably for various types of road surfaces, road types, speeds, and conditions. The algorithm relies on metrics such as heading angle, lateral position, radius of curvature, and peak yaw rate—calculated from yaw rate and vehicle speed data—to classify vehicle-movement events from a large file of vehicle data. Tuning parameters for the algorithm, set both analytically and empirically, are provided. The results of the tests show that the algorithm makes correct identifications of vehicle movement about 80% of the time. The highest success is for turns, followed by curves and lane changes. Although these numbers improve on previous methods, there is additional room for improvement that gives impetus for future work.