Trajectory Reconstruction at Multi-Lane Freeways Based on Fixed-Point Detector Data and Connected Vehicle Trajectory Data

Connected vehicle (CV) trajectory data provide a promising data source for the reconstruction of regular vehicle (RV) trajectories. However, most existing studies focus on the reconstruction of longitudinal trajectories of RVs and do not consider lane-changing (LC) behaviors, which is unrealistic. This study proposes a trajectory reconstruction model at multi-lane freeways considering both car-following (CF) and LC behaviors of RVs. The vehicle IDs, passing time, and passing speeds detected by fixed-point detectors and CV trajectory data are used. Taking two adjacent CVs in the same lane as a unit, the number of RVs within the unit is estimated by the hidden Markov model (HMM), which captures their LC probability. The forward-backward algorithm is designed for solutions. Further, the complete trajectories of RVs are reconstructed with the Helly CF model. Numerical research results show that (1) the accuracy of HMM is around 80% with the CV penetration rate of 20%, and (2) the proposed model outperforms the benchmark model by 11.6% in terms of the accuracy of trajectory reconstruction.