Probabilistic model for estimating vehicle trajectories using sparse mobile sensor data

Mobile sensors have emerged as a promising tool for traffic data collection and performance measurement, but most mobile sensor data today are sparse with low sampling rates, i.e., they are collected from a small subset of vehicles in the traffic stream every 10 to 60 seconds. Therefore, it is challenging to estimate the traffic states in both space and time based on these sparse mobile sensor data. In this paper, a stochastic model is proposed to estimate the second-by-second trajectories using sparse mobile sensor data. The proposed model investigates all possible driving mode sequences between data points. The likelihood of each scenario is quantified with mode-specific a priori distributions. Detailed trajectories are then reconstructed based on the optimal driving mode sequences. The proposed method is calibrated and validated using NGSIM data. It shows a 58.4% improvement on trajectory estimation, and a significant advance on mobility evaluation.