Performance evaluation of spatial indexing to identify polygon intersection

Abstract

This study evaluates the use of spatial indexes to identify polygon intersections as their results are often inaccurate. We analyzed the factors affecting performance and developed three types of indicators for evaluation. Four datasets with 4295, 53,646, 85,827 and 692,177 polygons were considered. Results show R-tree rapidly identified intersections, using less than 4% of the total time. However, it identified 1.8 or more times the actual number of polygon pairs and wasted over 43% of computation time calculating nonintersecting polygons. R-tree–based algorithm is more efficient for experimental datasets when the ratio of polygons in actually intersecting pairs to total polygons is below a threshold (4.88, 2.78, 4.66 and 3.87 with respect to the plane sweep algorithm). The threshold decreases when datasets contain polygons with larger mean number of vertices or complicated shapes and when computing intersections with lower efficiency. Compared with topology check, spatial indexing performs better with polygon overlay.