Open AccessSingle Image 3D Building Reconstruction Using Rectangles Parallel to an Axis
Author(s) -
Tomás Ferranti,
Asla Medeiros e Sá,
Paulo Cezar Pinto Carvalho
Publication year - 2021
Language(s) - English
Resource type - Conference proceedings
DOI - 10.5753/wvc.2021.18906
Subject(s) - computer science , computer vision , pinhole (optics) , computer graphics (images) , artificial intelligence , image (mathematics) , set (abstract data type) , visualization , schematic , 3d model , calibration , pinhole camera model , camera resectioning , mathematics , optics , camera auto calibration , engineering , statistics , physics , electronic engineering , programming language
Historic photographic collections are valuable documents of urban evolution through time. Many historic buildings documented in such collections may have been demolished or changed over time. Digital modeling such buildings may be challenging due to the reduced amount of information available that may be limited to a few images and/or schematic drawings. This paper presents a method to create a 3D set of rectangles that approximates elements of a scene (such as walls, floors, and roofs) from a single image. Using a pinhole camera model, the extraction of geometry and texture of planes parallel to an axis can be obtained after a camera calibration step that recovers intrinsic parameters of the model. Results show that a good visualization of the scene can be created, using the proposed technique, from a single image.