Design of spatial join index-based spatial join strategies

• Main Authors: Wu Ming-Jyh, 吳明智
• Other Authors: Chang Ye-In
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/86445001120821625584

碩士 === 國立中山大學 === 應用數學研究所 === 86 === Data stored in these spatial database systems includes simple geometric types like points, lines, polygons, and surfaces, and more complex types that are derived from the simpler geometric types. A spatial database system must support queries on these spatial objects efficiently. Spatial database users frequently need to combine two spatial inputs based on some spatial relationship between the objects in the two inputs. For example, map overlap, which requires combining two maps to produce a third, is an important operation in a spatial database. This operation of combining two inputs based on their spatial relationship is called a spatial join, which is an expensive operation. Moreover, the following key feature that makes the computation of spatial operators, more difficult than the computation of their non-spatial counterparts: There is no total ordering among spatial objects that preserves spatial proximity. Consequently, efficient spatial join algorithms are critical component of any spatial database system. Rotem's algorithm is a well-known approach for overlapoverlap detection, which is based on a new join index. However, there are some errors in Rotem's algorithm, which can result in some wrong answers. Therefore, we first correct the errors and then present a complete correct algorithm for overlap detection. Moreover, we extend Rotem's overlap detection algorithm for contain, cover, equal and meet detection. Brinkhoff et al. have proposed a different approach for overlap detection which sorts the entries in a node of the R-tree according to the spatial location of the corresponding rectangles. We also extend Brinkhoff et al.'s algorithm for cover, contain, equal and meet detection. Furthermore, we study the performance of both algorithms for overlap detection by simulation. From our simulation results, we show that Brinkhoff et al.'s algorithm needs longer join processing time, larger number of comparisons and a smaller number of buffers than Rotem's algorithm. Moreover, both the algorithms need sorting process first. For the index structure part, Rotem's algorithm creates join index, while Brinkhoff et al.'s algorithm is based on R-tree. Finally, we extend both algorithms to consider line to line relationships.