Processing Probabilistic K-Nearest Neighbor Query Using Rlsd-Tree

One of the important queries in spatio-temporal databases is the probabilistic K-nearest neighbor query (or PKNNQ for short). Given a query object q, a set S_mo of moving objects with uncertain moving speed and direction, a time point t_q, and a value of K, a PKNNQ finds the objects with highest probability of being the K-nearest neighbor of q at time t_q. Most of the existing approaches focus on designing an index structure for managing moving objects so as to facilitate processing the PKNNQ. However, they do not properly consider the moving speeds and directions of objects to build the index, and thus lead to poor query performance. In this paper, we use an efficient R^lsd-tree, which is built by applying space-filling curves to determine which objects are better to be grouped, to index moving objects. We develop pruning criteria combined with the R^lsd-tree to efficiently answer the PKNNQ. In addition, a probability model is designed to reasonably quantify the possibility of each object being the query result. Finally, a thorough experimental evaluation is conducted to show the merits of the proposed techniques.