Fuzing license plate recognition data and vehicle trajectory data for lane-based queue length estimation at signalized intersections

Abstract

Queue length is one of the most important performance measures for signalized intersections. With recent advancements in connected vehicles and intelligent mobility technologies, utilizing vehicle trajectory data to estimate queue length has received considerable attentions. However, most of the existing methods are based on some assumptions, such as known arrival patterns and/or high penetration rates. Besides, existing models would probably be unstable or invalid under sparse trajectory environment. Hence, license plate recognition (LPR) data is introduced in this study to fuze with the vehicle trajectory data, and then, a lane-based queue length estimation method is proposed. First, by matching the LPR data with probe vehicle data, the two-dimensional probability density distribution of discharge headway and stop-line crossing time of various kinds of vehicles, i.e., queued and nonqueued vehicle for undersaturated condition and twice-queued and once-queued vehicle for oversaturated condition, can be calibrated. Then, the Bayesian theory is adopted to derive the lane-based queue length for undersaturated condition as well as the initial queue for oversaturated condition with the largest possibility, respectively. Where probe vehicle trajectories, if existed, will provide the boundaries for the estimated queue lengths. Finally, the performance of the proposed method is evaluated using both simulation and empirical data. Simulation results show that the proposed method could produce accurate estimates of queue lengths for both undersaturated and oversaturated conditions and can achieve reliable estimates even under low penetration rate (3%). Empirical results show that the proposed method outperforms an existing method using probe vehicle trajectories only.