Towards automatic 3D reconstruction of pitched roofs in monocular satellite/aerial images

This thesis presents a novel method for estimating the three-dimensional shape of pitched roofs in monocular satellite/aerial images by utilizing the acquisition geometry (both sun and camera). This method consists of four steps: rooftop extraction, texture and vent removal, identification of planar sections, and 3D roof model generation. During the fourth step, 3D reconstruction candidates are partially rendered and iteratively compared against the original image to evaluate a fitness function. The experimental results illustrate that the proposed framework can derive 3D building rooftop models, including estimates for roofpitch, from single satellite/aerial images in situations where previous techniques fail.