Rural residential building extraction from laser scanning data and aerophotograph based on quadtree segmentation

A new method is presented for the extraction buildings in rural area from light detection and ranging (LIDAR) normalized digital surface model (nDSM) on the basis of quadtree segmentation principles. Compared with aerophotograph, LIDAR generally provides more accurate height information but less accurate boundary lines. In our method the two data sources are integrated for building boundary extraction. First, nDSMs are segmented to image objects by a quadtree Algorithm. Building candidates are obtained based on a certain height threshold of segmented objects of nDSMs. The brightness index derived from the aerophotograph is used to remove shadows. Then the spectral information is used to refine the edge of the building, which is implemented by classifying spectrally similar neighbors to the class of building in high steep area. The greenness index is extracted to eliminate the influence of vegetation. The trees over building can be reclassified by finding the image objects that are completely enclosed by other image objects belonging to building. At last, after two types of morphological operations and area threshold are applied to the building objects to eliminate the spurious objects and to smooth object boundaries, a set of reliable and clean building objects are obtained.