An integrated framework for fully sampled vehicle trajectory reconstruction using a fused dataset

Vehicle trajectories offer valuable insights for understanding traffic dynamics and optimising traffic control. However, the collection of fully-sampled vehicle trajectories is challenging due to unaffordable costs. To maximise the utility of sparse and limited trajectories, this study tailors an integrated framework for fully-sampled vehicle trajectory reconstruction. The proposed framework elaborates on a three-step work. Firstly, a piecewise cubic Hermite interpolating polynomial (PCHIP) is employed to reconstruct individual probe vehicle (PV) trajectories, and a piecewise order-changing model is proposed to capture overtaking dynamics. Secondly, a speed contour map is constructed to provide speed baselines for estimating undetected non-probe vehicle (NPV) trajectories on a region-by-region basis. Two candidate trajectories are estimated by conducting car-following (CF) model and inverse car-following (ICF) model, respectively. Thirdly, a weighted fusion model is designed to estimate NPV trajectories by integrating the model predictive control (MPC) algorithm. Comparative analysis proves that the combined model performs better than the pure CF model.