3D building reconstruction and construction site monitoring from RGB and TIR image sets

Abstract

This paper discusses methods on 3D reconstruction from ordered and unordered image sets in the visible RGB spectrum and the thermal infrared (TIR) spectrum. The 3D reconstruction is done detecting feature points in images and assigning the corresponding homologoues point. A bundle block adjustment with and without preknowledge of the camera orientations is used to coregister images of one image set. Next, dense matching is performed to calculate 3D object points for most of the image points. Resulting dense 3D point clouds are coregistered to other point clouds and polygonal building models. Facade textures are extracted from the images using corrected camera orientations of the 3D reconstruction. Change detection can be done based on 3D geometry and on radiometric changes like detected temperature changes in TIR based textures.