Real-Time Joint Estimation of Queue Dynamics and Arrival Time via a Car-Following-Model-Integrated Unscented Kalman Filter

Abstract

Optimizing trajectories of Connected and Automated Vehicles (CAVs) and signal plans requires precise real-time prediction of the Time of Arrival (ToA) at signalized intersections. Although CAVs feature advanced sensors that capture high-fidelity microscopic dynamics, their limited spatial coverage restricts the continuous observation of macroscopic queue profiles. To overcome this limitation, this paper proposes a real-time data fusion method integrating these mobile sensor data with a conventional stopline loop detector to jointly estimate queue length and predict ToA in mixed traffic environments. Specifically, the proposed method first initializes a macroscopic queue state using historical arrival patterns and data from stop line sensors. Subsequently, as the leader of the CAV decelerates, an Intelligent Driver Model-integrated Unscented Kalman Filter is triggered to continuously fuse the data of high-fidelity onboard sensors and shockwave dynamics. Under oversaturated conditions, the proposed method improves ToA prediction by reducing errors by 20.2% and 8.6% against pure macroscopic and moving-average constant deceleration baselines, respectively. By resolving the scale gap between macroscopic queue dynamics and microscopic car-following dynamics, this framework provides reliable predictions of queue length and ToA of CAVs.