Region based building footprint extraction and change detection for urban areas

Kurzfassung

Up-to-date geodata is important for a lot of applications like disaster management, or urban planning (Gamba and Houshmand, 2000). Thereby, the availability and quality of remote sensing based geodata increased largely in the last decades. Making use of “Big Data” has therefore become an important research topic. Especially the automatic extraction of building footprints and the detection of building changes has thereby a high scientific value and therefore many methods were proposed. These differ on the one side dependent on the used data. While some are using just photographic data, others are using height data in addition. In previous research there are thereby two main sources for the height data used. One is LIDAR (Light detection and ranging) - data, which has a very high accuracy. The other source is stereo imagery, were a DSM is calculated out of satellite stereo data. These stereo data based DSM are not as accurate as the LIDAR based DSM’s. However, they got the advantage of a better availability for large regions. For change detection, a major distinction can be seen between pixel and object based approaches. While pixel based approaches determining the building change for each pixel independently, object based approaches are determining the change based on groups of pixels. However, how to obtain accurate regions for building changes is still an open topic. Therefore, in thesis both building footprint detection and change detection is performed, based on satellite stereo data. Thereby a comparison is made between a just pixel based approach for change detection and methods that are object based. At first the state of art in terms of change detection and building footprint extraction is outlined. After that, the background for some of the basic functions is clarified. Next, the method explained in detail, before the results are described and the method is discussed.