Automatic 3D reconstruction and texture extraction for 3D building models from thermal infrared image sequences

This paper discusses the accuracies of two methods for automatic image orientation, 3D reconstruction, and extraction of textures of building facades from thermal IR image sequences. The first method uses the image sequence and camera calibration information only to reconstruct the scene in model coordinates and coregisters this model to a given 3D building model to derive optimized orientation parameters. The second method directly includes the given 3D building model into the bundle adjustment of the image sequence orientation. The resulting optimized orientation parameters are used to project the 3D model into the images and extract façade textures. Both methods are evaluated on their geometric accuracy using experimental image sequences.