Poster Abstract: Vehicular Mobility Modeling based on Heterogeneous Sensor Networks

Urban mobility especially vehicular mobility is important for many real-world applications. A lot of work has been done based on data from mobile sensor networks and stationary sensor networks, where each of them has its only weakness in spatiotemporal coverage and penetration rates. In this paper, we aim to coordinate the strength of heterogeneous sensor networks to achieve mobility modeling. Specifically, we design a mobility prediction system called Mohen to predict the destination of all the vehicles after they leave the coverage of stationary sensors with (i) only small portion vehicles with mobile sensors (i.e., GPS devices); (ii) historical data from stationary sensors (i.e., toll stations). The key novelty of Mohen is that we utilize the complementary features from heterogeneous sensor networks that mobile sensor networks provide better spatiotemporal coverage whereas stationary sensor networks capture vehicles with high penetration rates. We implement and evaluate Mohen in Guangdong Province, China by two real-world datasets, an electric toll collection (ETC) data set of 2 million vehicles, and an insurance-based vehicular tracking system of 114 thousand vehicles and show the accuracy of 75%.