A Hybrid building extraction algorithm based on multilevel segmentation and boundary simplification

Yarkınoğlu Gücük, Oya
Building footprints have an important role in urban planning, infrastructure development, climate studies and disaster management. The precise knowledge of buildings serves as a primary source for interpreting complex urban characteristics and also helps decision makers with more realistic urban management. In this thesis, the building boundary extraction including a hybrid approach is proposed. Firstly, normalized Digital Surface Model (nDSM) of the study area is generated from aerial Light Detection and Ranging (LiDAR) data. During the extraction methodology, multi-leveled segmentation is applied over nDSM data to detect possible building areas of differing sizes. Segmented regions of the first level are removed from nDSM and proceed to segmentation process until no regions exist. Finally accepted building footprint polygons are subjected to boundary extraction and line simplification steps to obtain smooth edges. Furthermore, proposed method results are compared with the results of Otsu algorithm and multi-threshold segmentation functionality of eCognition software. The building extraction accuracy of the proposed method is evaluated by considering pixel-based metrics. Finally, for selected 14 datasets the segmentation accuracy of the proposed method have better, detection rate and overall accuracy for 8 datasets. However, for 3 datasets accuracy measures of results are slightly different. For remaining 3 datasets Otsu algorithm gives better results.
Citation Formats
O. Yarkınoğlu Gücük, “A Hybrid building extraction algorithm based on multilevel segmentation and boundary simplification,” Ph.D. - Doctoral Program, Middle East Technical University, 2017.