Automatic Urban Modeling using Volumetric Reconstruction with Surface Graph Cuts

Ignacio Garcia-Dorado    Ilke Demir    Daniel G. Aliaga

Purdue University

Computers & Graphics 2013

Teaser image


The demand for 3D city-scale models has been significantly increased due to the proliferation of urban planning, city navigation, and virtual reality applications. We present an approach to automatically reconstruct buildings densely spanning a large urban area. Our method takes as input calibrated aerial images and available GIS meta-data. Our computational pipeline computes a per-building 2.5D volumetric reconstruction by exploiting photo-consistency where it is highly sampled amongst the aerial images. Our building surface graph cut method overcomes errors of occlusion, geometry, and calibration in order to stitch together aerial images and yield a visually coherent texture-mapped result. Our comparisons show similar quality to the manually modeled buildings of Google Earth, and show improvements over naive texture mapping and over spacecarving methods. We have tested our algorithms with a 12 square kilometer area of Boston, MA (USA), using 4667 images (i.e., 280GB of raw image data) and producing 1785 buildings.


Images & Video


Ignacio Garcia-Dorado, Ilke Demir, Daniel G. Aliaga. Automatic urban modeling using volumetric reconstruction with surface graph cuts. Computer & Graphics. Volume 37, Issue 7, November 2013, Pages 896–910.

 author = {Garcia-Dorado, Ignacio and Demir, Ilke and Aliaga, Daniel G.},
 title = {Automatic urban modeling using volumetric reconstruction with surface graph cuts},
 journal = {Computers & Graphics.},
 issue_date = {November 2013},
 volume = {37},
 number = {7},
 month = nov,
 year = {2013},
 issn = {0097-8493},
 pages = {896:1--910:15},
 numpages = {15},
 url = {},
 doi = {},
 publisher = {Elsevir},
 keywords = {automatic, urban, photo-consistency, graph cuts, volumetric reconstruction},