Fast Nearest Neighbor Search on Road Networks

Abstract

Abstract. Nearest neighbor (NN) queries have been extended from Euclidean spaces to road networks. Existing approaches are either based on Dijkstra-like network expansion or NN/distance precomputation. The former may cause an explosive number of node accesses for sparse datasets because all nodes closer than the NN to the query must be visited. The latter, e.g., the Voronoi Network Nearest Neighbor (V N 3 ) approach, can handle sparse datasets but is inappropriate for medium and dense datasets due to its high precomputation and storage overhead. In this paper, we propose a new approach that indexes the network topology based on a novel network reduction technique. It simplifies the network by replacing the graph topology with a set of interconnected tree-based structures called SPIE's. An nd index is developed for each SPIE and our new (k)NN search algorithms on an SPIE follow a predetermined tree path to avoid costly network expansion. By mathematical analysis and experimental results, our new approach is shown to be efficient and robust for various network topologies and data distributions.

Indexing Network Voronoi Diagrams

“…While IN E is an adaption of the Dijkstra algorithm, IER exploits the Euclidean restriction principle in which the results are first computed in Euclidean space and then refined by using the network distance. Several other kNN algorithms are proposed based on the improved (precomputation) version of INE [1,25,5]. In [19], Samet et al proposed shortest path quadtree algorithm for efficient evaluation of both shortest path and kNN queries in road networks.…”

Efficient Evaluation of k-Range Nearest Neighbor Queries in Road Networks

“…In location-based services, processing k-nearest neighbor (kNN) queries in the road network has been well studied (e.g., [3], [4], [5], [18], [19], [20], [21], [22], [23]). Among the existing solutions for kNN queries, the simplest one is the incremental network expansion (INE) algorithm [3].…”

“…However, the limitation of the INE algorithm is that it cannot take the advantage of the optimization based on the precomputed shortest distance. To overcome this limitation, there are many kNN query processing algorithms that utilize the precomputed network distance to optimize the query processing (e.g., [4], [5], [18], [19], [20], [21], [22], [23], [24], [25]). Although these optimized algorithms perform faster than the INE algorithm, they incur higher storage overhead.…”

“…INTRODUCTION k-Nearest Neighbor (or kNN for short) query is one of the most popular query types in location-based services [1], [2], where a user issues a kNN query to the service provider for the k-nearest objects of interest to her current location. With the advances in spatial databases, the kNN query has been extended from the Euclidean space to the road network environment [3], [4], [5], where the user can issue the kNN query to find her k-nearest objects of interest based on the network distance. The kNN query result over the network distance or the travel time is more meaningful to the user, since the user is usually traveling on the road network,.…”

Towards K-Nearest Neighbor Search in Time-Dependent Spatial Network Databases

“…Cho et al [2] presented a system UNICONS where the main idea is to integrate the precomputed kNNs into the Dijkstra algorithm. Hu et al [12] proposed a distance signature approach that precomputes the network distance between each data object and network vertex. The distance signatures are used to find a set of candidate results and Dijkstra is employed to compute their exact network distance.…”