Building extraction from UAV remote sensing data based on photogrammetry method

The accuracy of traditional spectral based land surface classification method can be decreased for areas with similarity spectral characters for different land cover types. To extract building distributions in those areas, a method based on the height information using unmanned aerial vehicle (UAV) optical remote sensing data were proposed in this study. With Digital Surface Model (DSM) and Digital Elevation Model (DEM) data produced by EasyUAV software using UAV acquired remote sensing images, the differences between DSM and DEM larger than 2.6 m was taken as a threshold for building extraction. It showed that the user accuracy and mapping accuracy of the building extraction method proposed in this study were 88.69% and 97.42%, respectively. To evaluate the performance of the proposed method, the buildings were also extracted based on traditional supervised classification method using Digital Orthophoto Map (DOM) produced by EasyUAV. The results showed that the new method was more accurate with the user accuracy and mapping accuracy of the supervised classification method of 43.23% and 85.30%, respectively.