VN-tree for efficient nearest neighbors search in location based services

Nearest neighbors queries are commonly used in location based services. This paper focuses on enhancing their processing based on indexing structures. Indexes, like R-trees, cannot avoid backtracking to process them. This feature weakens suitability of such an index to several data access modes within mobile environments. To meet needs of these environments, we propose a novel index called VN-tree (Voronoi-Neighbors tree), based on the Voronoi diagram and its dual graph (the Delaunay Triangulation) of the candidate points of interest. VN-tree adopts a non-overlapping partitioning scheme and supports backtracking-free search algorithms to answer several types of queries. In particular, processing continuous nearest neighbor queries is a straightforward use of VN-tree. We conduct various experiments to compare VN-tree with R*-tree as in-memory indexes in the on-demand data access mode. Main results show that VN-tree consistently provides outstanding performances.