A Bayesian Estimation Approach for Spatiotemporal Vehicle Trajectory Reconstruction Using Multi-Source Traffic Flow Data

Vehicle trajectory is an emerging data source used in various applications. However, only a few vehicles can provide detailed trajectories, leaving the population vehicle trajectories unknown. Previous studies have successfully estimated macroscopic traffic states but may not estimate detailed vehicle trajectories directly. In this study, Newell’s simplified kinematic wave model is used to solve the vehicle trajectory reconstruction problem. A Bayesian-based approach is proposed to estimate the model parameters using the cumulative flow data collected at the link boundaries and sparse vehicle trajectory data collected on the link. The calibrated model is used solely to reconstruct entire vehicle trajectories on the link using link boundary flow detection. The proposed method is validated with the NGSIM data set on an urban road. The mean average error of the reconstructed vehicle trajectory is about 8 m. The impacts of cumulative flow and trajectory point errors are also investigated.