Advances on K nearest neighbour search in spatial databases

Abstract

A spatial database is a database that stores data and makes queries which are related to objects in space, including points, lines and polygons. The spatial database is designed to process the spatial data type which cannot be processed by typical databases. $k$ Nearest Neighbor search is a type of query to classify objects based on closest distances in the feature space, as a result, a large portion of $k$ nearest neighbor search queries are based on road network. Consequently, the most popular method that has been fully discussed is called Network Expansion. By processing more and more complex queries, network expansion methods show a significant drawback which is poor performance because the underlay road network is crossed and connected. Do expansion for each intersection points in road network is unrealistic. The other problem in $k$ Nearest Neighbor ($k$NN) query processing is that most of the existing $k$ nearest neighbor searches are concentrated on points. In other words, the discrete points are the input and output of $k$ Nearest Neighbor search query. While in reality, points, lines as well as polygons are three types of spatial elements. How to utilize Lines/Route in spatial query becomes another question for our researchers. Motivated by these two points, the following paragraph summarizes the contribution of this thesis. The first main contribution of this thesis is called emph{Voronoi Based $k$ Nearest Neighbor search query}. Compared to the Network Expansion method, emph{the Voronoi based method} aggregates the road segments using a Voronoi Diagram. Instead of expanding from each intersection in road network, the Voronoi Diagram divides the map into polygons by treating the points of interest as generators. In this part, we have proposed 2 algorithms which use the Voronoi Diagram to improve the performance for Multiple types of $k$ Nearest Neighbor Search query and Continuous $k$ Nearest Neighbor Search query respectively. Our experiments have shown the proposed algorithms can improve the performance significantly either from a cost and a processing time point of view compared to the traditional Network Expansion method. The second main contribution of this thesis is called emph{Route and Path related $k$NN Queries}. The aim of this part in the thesis is to bring emph{lines/routes} into the input and(or) output of $k$ Nearest Neighbor search. As a result, users can use lines to find points, use points to find routes, or even use route to find route. Correspondingly, there are three novel approaches discussed in this part, namely, Path Based $k$NN Search Queries, Path Branch Point based $k$NN Search Queries and Time Constraint Route Search over Multiple Locations. By proposing these three approaches, the $k$ Nearest Neighbor Search has been enriched and can satisfy various types of user queries. To sum up, the thesis is concentrated on two main category of query processing: i) emph{Voronoi based $k$ Nearest Neighbor Search} which is aiming at improving the Network Expansion Method which is the most popular technique used by existing $k$ Nearest Neighbor Search approaches. ii) emph{Route and Path related $k$NN Query} which brings route and path into the input or/and output of $k$NN queries and which fills up the blanket area in $k$NN query that only concrete points are utilized in spatial queries.