Calibrating Car-following Models for Urban Traffic Using Top-view Trajectory Data

Abstract

Microscopic traffic simulation models are useful tools for decision support systems to evaluate transportation scenarios and transportation system performances. However, parameter identifications of microscopic driving models are difficult because the behavior parameters are generally not directly observable from the conventional data collection approach. This paper proposes a novel method for calibrating and validating car-following models using top-view trajectory data. The vehicle trajectory data is high-fidelity and provides detailed information on the driving characteristics. The trajectory data were extracted using deep-learning-based computer vision technologies from aerial video footage collected using an unmanned aerial vehicle (UAV). The methodology is applied for the parameter estimation with PTV VISSIM using 2,871 trajectories obtained from an intersection in Taiwan. For the validation, trajectory-level measures of performances are also proposed, including speed distribution, acceleration and deceleration distribution, and discharge headway. The results indicate that the realism of traffic simulation models is improved.