Fusing heterogeneous traffic data by Kalman filters and Gaussian mixture models

Heterogeneous traffic data collected from different types of sensors are fused for estimating traffic states more accurately. Data quality and fusion method are two key issues required to be solved in the traffic state estimation. In this paper, we propose a fusion method of heterogeneous traffic data based on the Kalman filters (KF) and Gaussian mixture models (GMM). The noise in collected raw data is reduced by the KF in order to improve the quality of input data for fusion. The vectors of historical data from global positioning system (GPS) and remote traffic microwave sensors (RTMS) in different traffic states are modeled with multiple multi-variate GMM respectively. Finally, the estimated traffic state can be obtained by computing the posterior probabilities with the vector data and GMM. Performance of our work is examined by series of experiments, and the results show that the proposed method is effective for improving the precision of traffic state estimation.