Automatic Building Detection Using the Dempster-Shafer Algorithm

An approach and strategy for automatic detection of buildings from aerial images using combined image analysis and interpretation techniques is described in this paper. It is undertaken in several steps. A dense DSM is obtained by stereo image matching and then the results of multi-band classification, the DSM, and Normalized Difference Vegetation Index (NDVI) are used to reveal preliminary building interest areas. From these areas, a shape modeling algorithm has been used to precisely delineate their boundaries. The Dempster-Shafer data fusion technique is then applied to detect buildings from the combination of three data sources by a statistically-based classification. A number of test areas, which include buildings of different sizes, shape, and roof color have been investigated. The tests are encouraging and demonstrate that all processes in this system are important for effective building detection. Introduction One of the major challenges in the fields of computer vision and digital photogrammetry is the 3D reconstruction of the terrain surface from aerial images of urban or suburban areas where buildings, roads, trees and vegetation are intermingled in an intricate and complex fashion. The automatic determination of Digital Terrain Models (DTM) by stereo image matching algorithms has been one of the primary goals of digital photogrammetry for many years, particularly for the production of digital orthophotos, 3D building reconstruction, 3D city models, the application and management of 3D databases for urban and town planning, and Geographic Information Systems (GIS) modeling. Stereo image matching determines corresponding pixels or features in overlapping images and is fundamental to digital photogrammetry for elevation determination. However, conventional image matching supplies a Digital Surface Model (DSM) or visible surface, since it determines elevations of the tops of man-made objects such as buildings, or vegetation, and hence does not represent the terrain surface (Baltsavias et al., 1995; Henricsson et al., 1996; Tönjes, Automatic Building Detection Using the Dempster-Shafer Algorithm Yi Hui Lu, John C. Trinder, and Kurt Kubik 1996). Therefore, it is necessary to identify buildings, trees and other objects on the surface to be able to reduce the elevations to the bare earth DEM. Many automated building detection and extraction methods have been proposed by researchers. Shadow analysis-based algorithms have been used by Liow and Pavlidis (1990) and Nevatia et al. (1999). Information fusion-based systems have been reported by McKeown (1991), and Haala and Hahn (1995). Methods supported by DTM and orthoimages have also been reported by Baltsavias et al. (1995), Horiguchi et al. (2000), Straub and Heipke (2001), and Brunn (2001). Considering the different shapes, environments, and image intensity for different buildings, together with the occurrence of occlusions and shadow effects, the automation of building extraction is a complicated and difficult procedure (Sahar and Krupnik, 1999). In addition to developing better schemes, the inclusion of more information is an essential direction for the research. Henricsson (1998), Chen and Hsu (2000), and Niederost (2001) used color images to improve the system performance for roof determination and edge extraction. Spreeuwears et al. (1997) and Gabet et al. (1997) used multi-view images to reduce the effect of occlusions. Multiimage 3D feature and DSM extraction for building change detection were proposed by Paparoditis et al. (1998 and 2001). Laser scanner data were used by Mcintosh et al. (2000), Masaharu and Hasegawa (2000), and Haala et al. (1998). Multi-resolution analysis of wavelets for house extraction has been proposed by Shi and Shibasaki (1995). Huertas and Nevatia (1988), Shufel and McKeown (1993), Henricsson et al. (1997), Lammi (1996), Henricsson and Baltsvias (1997), Jaynes et al. (1997a), Jaynes et al. (1997b), Baillard et al. (1998), Kim and Muller (1998), Vosselman (1999), and Lu et al. (2003) have also demonstrated developments in algorithms for building extraction. Despite the development by researchers of many automatic building extraction algorithms based on images, terrain laser scan data and their combination (Collin et al. 1998; Hanson et al. 2001; Henricsson, 1998; Walter, 1999; Haala and Brenner, 1999), there are no operational algorithms because each method is focused on a particular application and data sources, and is usually not transferable to different features. The goal of this research is to define building areas occurring on overlapping aerial or satellite images over a variety of terrain types and ground cover, for the reconstruction of terrain elevations of the bare earth surface, without the input of additional data such as terrain laser scanning or GIS. The approach includes an attempt to understand and PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Ap r i l 2006 395 Yi Hui Lu is with the NSW Department of Environment and Conservation, Sydney 2000 NSW, Australia (Yi.Lu@ environment.nsw.gov.au) She was formerly at the School of Surveying and SIS, University of NSW, UNSW Sydney 2052 NSW, Australia. John C. Trinder is with the School of Surveying and SIS, University of NSW, UNSW Sydney 2052 NSW Australia ([email protected]). Kurt Kubik is with the Department of Computer Science and Electrical Engineering, University of Queensland, QLD 4072 Australia ([email protected]). Photogrammetric Engineering & Remote Sensing Vol. 72, No. 4, April 2006, pp. 395–403. 0099-1112/06/7204–0395/$3.00/0 © 2006 American Society for Photogrammetry and Remote Sensing 04-118 3/14/06 9:04 PM Page 395