Traffic-Known Urban Vehicular Route Prediction Based on Partial Mobility Patterns

Travel route analysis and prediction are essential for the success of many applications in Vehicular Ad-hoc Networks (VANETs). Yet it is quit challenging to make accuracy route prediction for general vehicles in urban settings due to several practical issues such as very complicated traffic networks, the highly dynamic real-time traffic conditions and their interaction with drivers' route selections. In this paper, we undertake a systematic study on the vehicular route prediction in urban environments where the traffic conditions on complicated road networks keep changing from time to time. Inspired by the observation that a vehicle often has its own route selection flavor when traversing between its sources and destinations, we define a mobility pattern as a consecutive series of road segment selections that exhibit frequent appearance along all the itineraries of the vehicle. We further leverage Variable-order Markov Models (VMMs) to mine mobility patterns from the real taxi GPS trace data collected in Shanghai. In addition, considering the tremendous impact of dynamic traffic conditions to the accuracy of route prediction, we deploy multiple VMMs differentiating different traffic conditions in daytime. Our extensive trace-driven simulation results show that notable patterns can be mined from routes of common vehicles though they usually have no constraints when selecting routes. Given a specific taxi, around 40% next road segments are predictable using our model with a confidence weight of 60%. With multiple VMMs a high route prediction accuracy is achievable from the real traffic trace.