Building Roof Component Extraction from Panchromatic Satellite Images Using a Clustering-Based Method
Keywords: Pan-chromatic image, DSM, Roof Type Detection, Mask Refinement, Clustering
Abstract. Developing fully automatic systems is still an active research topic in 3D building model reconstruction. While a general solution to the building reconstruction problem relies on collecting and grouping the modeling cues (e.g., lines, corners, planes) from Digital Surface Model (DSM) data, failure in finding the cues due to noise in the DSM and the object complexities is a big challenge. In this paper, we introduce a clustering-based method for cue discovery from Pan-chromatic satellite images which reduces the dependencies of the reconstruction techniques on DSM data. Experimental results show that the proposed method is not only able to effectively refine building masks by discriminating building boundaries from nearby clutter, but also is able to determine the roof types (e.g., pitched, flat). The latter, allows to establish a reconstruction method to reduces the search effort and the failure probability regions in finding a particular cue by leading the system to an appropriate area.