A data fusion framework is proposed to predict vehicle trajectories in urban signalized arterials. Recent advancements in data collection techniques and availability of traffic data from fixed and probe sensors suggest data fusion-based models as an alternative approach for traffic prediction. Yet, majority of the existing data fusion approaches are based on statistical models without considerations for traffic engineering principles. In addition, existing approaches do not use probe trajectory data efficiently. The proposed framework in this research is based on the kinematic wave theory and is capable of fully utilizing the probe trajectory data to reconstruct the trajectories of the other vehicles within a 'prediction window'. The data fusion framework combines real-time and historical traffic data to predict future traffic patterns at upstream and downstream boundaries. The modelling approach is based on the kinematic wave theory and applies the variational theory as the solution method. Predicted traffic patterns are used to set the boundary conditions in the solution domain and probe trajectories are used to set additional boundary conditions. Given the boundary conditions, a dynamic programming approach is applied to reconstruct vehicle trajectories within the prediction window. The performance of the proposed framework is evaluated by using real-world traffic data, and possible directions for improving the accuracy of the model are discussed.
ASJC Scopus subject areas