Efficient Processing of Large Spatial Queries Using Interior Approximations

Spatial data in CAD/CAM and geographic information systems involve arbitrarily-shaped 2and 3-dimensional geometries. Queries on such complex geometry data involve identification of data geometries that interact with a specified query geometry. Since geometry-geometry comparisons are expensive due to the large sizes of the data geometries, spatial engines avoid unnecessary comparisons by first comparing the MBRs and filtering out irrelevant geometries. If the query geometry is large compared to the data geometries, this filtering technique may not be effective in improving the performance. In this paper, we describe how to reduce geometry-geometry comparisons by first filtering using the interior approximations of geometries (in addition to and after comparing the exteriors, i.e., the MBRs). We implemented this technique as part of the R-tree indexes in Oracle Spatial and observed that the query performance improves by more than 50% (or a factor of 2) for most queries on real spatial datasets.