Using stroke and mesh to recognize building group patterns

Abstract

Building patterns are crucial structures and should be preserved in map generalization. However, while building patterns are not explicitly described in building datasets, map readers perceive building patterns effortlessly. Hence, to better support map generalization, it is important to automatically recognize building patterns in such datasets. This paper first proposes an extended and integrated typology of different building patterns. Based on the typology, building patterns are recognized using stroke and mesh. This method first structures the proximity graph of buildings, and then introduces six constraints (distance, size, shape, orientation, elongation, and facing ratio) to refine the original proximity graph. Strokes and meshes are derived from the refined proximity graph, and are used to recognize linear and grid building patterns, respectively. The proposed method is tested in four regions that are representative of different pattern types. The recognition results are evaluated in an expert survey and compared with the minimum spanning tree method. Assessment suggests that the linear and grid patterns in suburban and rural areas are recognized with satisfying results.