Big-Data Approach for Three-Dimensional Building Extraction from Aerial Laser Scanning

This paper proposes a big-data approach to automatically identify and extract buildings from a digital surface model created from aerial laser scanning data. The approach consists of two steps. The first step is a MapReduce process where neighboring points in a digital surface model are mapped into cubes. The second step uses a non-MapReduce algorithm first to remove trees and other obstructions and then to extract adjacent cubes. According to this approach, all adjacent cubes belong to the same object and an object is a set of adjacent cubes that belong to one or more adjacent buildings. Finally, an evaluation is presented for a section of Dublin, Ireland, to demonstrate the applicability of the approach, resulting in a 91% quality level for the extraction of 106 buildings over 1 km2, including buildings that have more than 10 adjacent components of different heights and complicated roof geometries. The proposed approach is notable not only for its big-data context but for its usage of vector data.