STC: Coarse-Grained Vehicular Data Based Travel Speed Sensing by Leveraging Spatial-Temporal Correlation.

As an important information for traffic condition evaluation, trip planning, transportation management, etc., average travel speed for a road means the average speed of vehicles travelling through this road in a given time duration. Traditional ways for collecting travel-speed oriented traffic data always depend on dedicated sensors and supporting infrastructures, and are therefore financial costly. Differently, vehicular crowdsensing as an infrastructure-free way, can be used to collect data including real-time locations and velocities of vehicles for road travel speed estimation, which is a quite low-cost way. However, vehicular crowdsensing data is always coarse-grained. This coarseness can lead to the incompleteness of travel speeds. Aiming to handle this problem as well as estimate travel speed accurately, in this paper, we propose an approach named STC that exploits the spatial-temporal correlation among travel speeds for roads by introducing the time-lagged cross correlation function. The time lagging factor describes the time consumption of traffic feature diffusion along roads. To properly calculate cross correlation, we novelly make the determination of the time lagging factor self-adaptive by recording the locations of vehicles at different roads. Then, utilizing the local stationarity of cross correlation, we further reduce the problem of single-road travel speed vacancy completion to a minimization problem. Finally, we fill all the vacancies of travel speed for roads in a recursive way using the geometric structure of road net. Elaborate experiments based on real taxi trace data show that STC can settle the incompleteness problem of vehicle crowdsensing data based travel speed estimation and ensure the accuracy of estimated travel speed better, in comparison with representative existing methods such as KNN, Kriging and ARIMA.