Traffic State Interpolation of Unobserved Road Links by Accumulated Data Using the Relationships of Links in Road Network

Abstract

For road traffic monitoring, probe vehicles data is useful. However, road link speed data observed by probe vehicles can include spatial and temporal missing data. To solve the problem, we model the traffic states in road network by Gaussian Graphical Model (GGM) and interpolate the road link speed. It is the difficult problem to estimate the parameters of multivariate normal distribution because of parameters’ high dimensionality. We propose two approaches, one is structural GGM and another is the estimation by Graphical Lasso and we extend these methods for partial observation by EM algorithm. Finally, we evaluate the performance of our method using one month probe data in central Bangkok from the aspect of interpolation accuracy and computational time and the result indicates that our approach is higher interpolation accuracy and shorter computational time.