An Approximate Indexing Method for Efficient Processing of k-Nearest Neighbor Queries in Road Network Environment

Abstract

In this paper; we address approximate indexing for efficient processing of k-nearest neighbor (k-NN) queries in road network databases. Previous methods suffer from either serious performance degradation in query processing or large storage overhead because they did not employ indexing mechanisms of objects based on their network distances. To overcome these drawbacks, this paper proposes a novel method that builds an index on those objects on a road network by approximating their network distances and processes k-NN queries efficiently by using that index. We first propose a way of mapping each object on a road network into the corresponding point in m-dimensional Euclidean space. For this, we propose using a new notion of an average network distance between arbitrary two objects, and prove that this distance satisfies the three properties of symmetry, reflexivity and triangular inequality required for indexing. Second, we propose an approximate indexing algorithm that builds an R*-tree on objects on a road network based on the mapping mechanism. Third, we propose a query processing algorithm that performs k-NN queries efficiently using the index. Finally, we verify the accuracy of the proposed method via extensive experiments using the real-life road network databases.