Yüksek Yoğunluklu Kentsel Alanlarda Çok Yüksek Çözünürlüklü Gerçek Ortofotolardan Kural Tabanlı Sınıflandırma Yöntemi İle Bina Çıkarımı

Abstract

The importance of aerial imagery is increasing day by day in terms of evaluating the current situation, following the construction and determining the direction of development of the cities. Urban transformation projects are also an important basis for these buildings are up to date, fast and reliable information is needed. With the advancement of technology, it is used extensively in the process of removing buildings from urban areas where population density is high, planning urban development, preventing illegal constructions, monitoring building changes, conducting municipal services properly and implementing large projects from the aerial images whose usage areas are widening. In this thesis, rule based automatic building extraction is aimed by using 10 cm resolution Orthophoto Image and Digital Surface Model (DSM) for four test areas with different structural characteristics in Altındağ, Keçiören and Etimesgut districts of Ankara. Building extraction was carried out based on spatial observations by identifying various sets of rules that complement each other. The main steps of the rule based approach used are as follows: First, ground filtering was performed on DSM data and Digital Terrain Model (DTM) data was generated. The Normalized Digital Surface Model (nDSM) was obtained by subtracting the bare land surface (DTM) produced by ground filtering from DSM. A 5x5 size median filter was applied on the orthophoto to reduce local variations, to eliminate noise and to clarify the edges of objects. Afterwards, multi-resolution segmentation was performed on the filtered orthophoto image, the most appropriate scale parameter was determined using the ESP-2 software and the homogeneous image objects were generated. Then, for the purpose of automatic building extraction, various sets of rules have been defined in accordance with the spectral and structural properties of the acquired image objects. Accuracy analysis was performed by comparing the building data obtained from the building extraction process with the vector reference data. As a result of the analysis, 89.65% accuracy, 87.81% accuracy, 91.47% accuracy in Test Area-2, 94.67% accuracy in Test Area-1, 94.37% accuracy in Test Area-3, 91.70% integrity, 88.41% accuracy in Test Area-4, 90.56% integrity values were reached.The results show that the method developed within the scope of the study is very successful in the extraction of buildings from high-resolution colored orthophotos.