Building extraction using local surface normal angle transformation

摘要

Building detection and extraction from digital surface model (DSM) becomes more and more attracted when high resolution airborn radar and CCD sensor find their more applications in photogrammetric field. Because DSM contains the terrain heights, we firstly derive DTM from DSM, and then generate Normalized DSM (NDSM). The buildings are extracted from NDSM. However, since urban appearance is complex, with large buildings, small buildings, trees in mass, etc., building extraction is implemented through several stages. Local Surface Normal Angle Transform (LSNAT) is implemented to the height field. Big buildings are distinguished from other large regions generated from binary NDSM. Watershed segmentation is used to detect trees in mass together with LSNAT. Roofs of the small building are extracted based on the histogram of LSNAT. A case study is presented and analyzed in the end of this paper.