Vehicle Trajectory Estimation based on Dynamic Bayesian Networks

In this paper we propose a method to estimate the most likely region a vehicle will transverse when the current position and a priori spatial-temporal trajectory data of multiple vehicles are known. The proposed solution is based on a hidden Markov chain that models the trajectory of each vehicle. The estimation of the vehicle trajectory relies on Viterbi algorithm, which identifies the most likely trajectory as a new vehicle's location is known. The proposed modeling and estimation methodology is evaluated using real mobility traces sampled from multiple taxis traveling in the city of Porto in Portugal. The estimation performance of the most likely region a vehicle will cross is between 32% and 89%, depending on several factors that include the probability of trajectory's occurrence and the number of previous locations considered in the estimation methodology (Markov order). Finally, we discuss the advantages and limitations of the proposed method.