Predictive line queries for traffic prediction

The advances in communication and positioning device technologies have made it possible to track the locations of moving objects, such as vehicles equipped with GPS. As a result, a new series of applications and services have been commenced into people's life. One popular application is the real-time traffic system which provides current road condition and traffic jam information to commuters. To further enhance this location-based experience, this paper proposes an advanced type of service which can predict traffic jams so that commuters can plan their trips more effectively. In particular, traffic prediction is realized by a new type of query, termed as the predictive line query, which estimates the amount of vehicles entering a querying road segment at a specified future timestamp and helps query issuers adjust their travel plans in a timely manner. Only a handful of existing work can efficiently and effectively handle such queries since most methods are designed for objects moving freely in the Euclidean space instead of under road-network constraints. Taking the road network topology and object moving patterns into account, we propose a hybrid index structure, the RD-tree, which employs an R*-tree for network indexing and direction-based hash tables for managing vehicles. We also develop a ring-query-based algorithm to answer the predictive line query. We have conducted an extensive experimental study which demonstrates that our approach significantly outperforms existing work in terms of both accuracy and time efficiency.