Traffic Speed Estimation Based on Multi-Source GPS Data and Mixture Model
Pu Wang,Zhiren Huang,Jiyu Lai,Zhihao Zheng,Yang Liu,Tao Lin
DOI: https://doi.org/10.1109/tits.2021.3095408
IF: 8.5
2022-01-01
IEEE Transactions on Intelligent Transportation Systems
Abstract:The traffic speed information of an urban road network is generally estimated using the widely available taxi GPS data. However, taxi usages are preponderantly restricted to areas with high population density, which results in limited spatial coverage of collected taxi GPS data. Moreover, the traffic speeds of taxies are not guaranteed to well represent the traffic speeds of other types of vehicles. In this study, we address these issues by introducing an infinite Gaussian mixture model to estimate traffic speed distribution. The variational inference method is employed to deal with the complicated parameter estimation problem. The proposed mixture model simultaneously combines taxi GPS data, bus GPS data, and mobile phone GPS data, which not only generates the mixed traffic-speed distribution of different types of vehicles but also improves the spatial coverage and the quality of traffic speed estimation. Surprisingly, we find that the incorporation of mobile phone GPS data can considerably improve the model’s ability to sense anomalous traffic conditions. Finally, the mixed traffic-speed distribution is validated using the license plate recognition data.