Stochastic queue profile estimation using license plate recognition data

Abstract

Queue profile has garnered significant attention for its precise description of queue formation and dissipation at signalized intersections. License Plate Recognition (LPR) systems, recording full-sample lane-based departure times of individual vehicles, offer an ideal data source for queue profile estimation. Thus, this paper presents a stochastic queue profile estimation method utilizing LPR data. It adopts the classic Input-Output model as a baseline to reconstruct arrival and departure curves using departure information from LPR detectors at successive intersections. A pseudo departure curve is then derived from the distribution of free-flow travel time. The stochastic queue profile is further reconstructed by analyzing the relationship among these curves, accounting for potential lane changes and overtaking behavior. The proposed method is validated through empirical and simulation cases. The empirical case achieves accurate queue length estimation with a Mean Absolute Error (MAE) of 5.38 m, while the simulation case yields precise stochastic queue profiles with an MAE of 5.95 m and an average deviation of 2.40 m for queue length. Sensitivity analysis demonstrates the method's robustness across parameters like demand-to-capacity ratio, miss detection ratio, and overtaking ratio. This work holds promise for real-time signal control optimization and vehicular trajectory reconstruction.