Three-dimensional reconstruction of urban scenes from a single image

Abstract

This work presents a method to output the 3D structure of an urban area from only one aerial photo. The reconstruction is based on the inverse process done when a parallel projection is used to construct a two-dimensional image from a three-dimensional object. A new algorithm for reconstruction is developed and implemented. It recovers depth information by identification of object height on the image plane. While the 3D-objects are reconstructed, their original texture is used for rendering, using control points defined by the user. Results show that when seen from the same capture angle, the virtual 3D scene and the original picture are exactly the same. In the virtual environment of this implementation, the user can visualize, visit and interact with the structure of the 3D scenario. The spatial structure of the real world is recovered at real time; it is possible to see the 3D growing out of the 2D images.