A Study on Spatial Query Processing in Euclidean Space

• Main Authors: Huang, Chen-Che, 黃振哲
• Other Authors: Huang, Jiun-Long
• Format: Others
• Language: en_US
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/50507146065611962812

博士 === 國立交通大學 === 資訊科學與工程研究所 === 101 === With the proliferation of location-aware mobile devices and the extensive coverage of wireless networks, location-based services (LBSs) have been rapidly gaining in popularity in recent years. Spatial queries, one of LBSs, are to retrieve the data objects of interest according to the locations of query users. In addition to being one useful LBS, spatial queries can be applied to a wide variety of applications such as profile-based marketing, decision support, and resource allocation. Due to a broad spectrum of applications of spatial queries, a significant number of research studies on spatial queries have been proposed and could be broadly classified into two categories, namely traditional spatial query exploration and new spatial query introduction. In this dissertation, we propose two works where one belongs to traditional spatial query exploration and the other new spatial query introduction. In the first work, we propose a proxy-based approach to continuous spatial queries in mobile environments. The proposed approach aims to efficiently provide effective estimated valid regions (EVRs) of NN and window queries for mobile clients to reduce the numbers of submitted queries as well as the load on the query processing server. For NN queries, we devise novel algorithms to allow the proxy to immediately offer effective EVRs to mobile users. The algorithms also enable the proxy to build effective EVRs even when the proxy cache size is small. On the other hand, for window queries, we propose to represent the EVRs of window queries in the vector form, called estimated window vectors (EWVs), to achieve larger estimated valid regions. Besides, a companion algorithm is presented to create the EWVs in an efficient manner. Due to the distinct characteristics of NN and window queries, we employ different index structures, namely an EVR-tree for NN queries and a grid index for window queries. Moreover, to further increase the effectiveness and efficiency of the proposed approach, we develop algorithms to make the EVR-tree and the grid index mutually support each other. Specifically, the grid index is used to answer NN queries and extend existing EVRs. The answer objects of NN queries are exploited to update the grid index, benefiting the creation of more effective EWVs of window queries. The extensive experimental results demonstrate that the proposed approach significantly outperforms the existing proxy-based approaches in terms of client cache hit ratio, query answering time, and server load. Additionally, the results indicate that the performance of the proposed proxy-based approach is close to that of the representative server-based approach even though the proposed approach has only partial information of data objects. In the second work, we propose and investigate a novel type of spatial queries,namely nearest dense window (NDW) queries. An NDW query $(q,l,w,k)$, a variant of NN queries, retrieves the nearest dense window of length $l$ and width $w$ containing at least $k$ data objects inside the window with respect to the query location $q$. The rationale of introducing the NDW query is that a region with a number of data objects is able to allow a query user to explore more objects and to be more likely to find satisfactory objects at a low moving cost. To answer NDW queries, we introduce and identify several properties to facilitate the identification of dense windows. Based on the properties, we develop a baseline algorithm to find the nearest dense window of an NDW query. Although the baseline algorithm can be used to answer NDW queries, we propose an NDW algorithm with less computation and I/O access to further increase the efficiency of NDW query search. The NDW algorithm mainly consists of three techniques, namely dense window discovery, search region reduction, and index node pruning. The dense window discovery technique attempts to avoid evaluating irreverent objects and to terminate the discovery as early as possible. The search region reduction and index node pruning techniques take advantage of the best distance found so far to reduce the search cost of objects and index nodes, respectively. The experimental results show that the proposed NDW algorithm is efficient in terms of I/O cost and query cost.