Integrated Method of Building Extraction from Digital Surface Model and Imagery

In urban GIS and management, Digital surface model (DSM) can be used for generating object models. The preliminary step is to segment the interesting areas. In this paper, we present an integrated method of building extraction using transform from DSM to normal angles and watershed segmentation to the gradient of DSM. To remove the effect of the terrain shape to building detection, we firstly derive DTM from DSM by geomorphology filtering, and then generate Normalized DSM (NDSM) by subtracting DTM from DSM. Then the buildings can be extracted from NDSM. Since urban appearance is complicated with large buildings, small buildings, and woods, etc., building extraction is implemented through several stages. In the first stage, Local Surface Normal Angle Transformation (LSNAT) is implemented to DSM to extract lant roof buildings. In the second stage, Marker based watershed is implemented to get the boundaries of the objects above the ground. The result of marker based buildings and LSNAT based buildings are merged to improve the building extraction accuracy. At last, the orthogonal imagery is used to remove the woods according to the green color principle. A case study is implemented to a DSM of a region of Kounan, Japan with spatial resolution of 1m.