3D Reconstruction of Indoor and Outdoor Building Scenes From a Single Image

In this paper, a novel method is proposed able to automatically generate accurate 3D models of both outdoor buildings and indoor scenes with perspective cues from line segments that are automatically extracted from a single image with an un-calibrated camera. The proposed method uses geometric constraints and knowledge of photography and achieves an accurate, real-time and fully automated 3D reconstruction of the scene without any intervention from the user. Image-based modeling relies on a set of techniques for creating a 3D representation of a scene from one or more 2D images. Although humans easily infer depth from images, computer and robots using low-level image processing perform more poorly.