A probabilistic approach for traffic state estimation

Traffic state estimation plays a significant role in intelligent transportation systems. For a specific road, traffic state varies at different times of a day. Therefore, to estimate real-time traffic state is difficult. This paper presents a probabilistic approach for traffic state estimation. In this approach, traffic state distribution and data point distribution are used to describe the pattern of the traffic state. By combining these two distributions, a probabilistic traffic state model can be derived for a specific road. Using multiple pairs of distribution, the change of traffic state in one day can be described. The parameters in this model can be estimated from traffic data using Gibbs sampling algorithm, and once the model is determined, the traffic state can be estimated from the real-time traffic data collected from traffic detectors. We conducted an experiment using traffic data from Haining City, and the result shows the feasibility and expansibility of this probabilistic approach.