Texture extraction for building models from IR sequences of urban areas

Focus of the paper lies on automated texturing of 3d building models with images recorded with infrared (IR) cameras. Therefore, a data fusion of given 3d models provided by a GIS and recorded IR image sequences is performed. Two concepts are presented to project the 2d images into the 3d object space. In the first concept, the polygones of the faces of the 3d model are projected into the IR image and matched to the edges for assigning the correct parts of the image to the corresponding faces of the model. These image parts are then projected onto the faces and stored as surface textures. The second concept uses homography to detect surface planes in two subsequent images and matches theses planes to the 3d model. Because one image does not show complete building facades, the extracted textures of the images are combined to create complete textures for the model surfaces. These textures can be used for feature extraction and object recognition for analyzing buildings.