Premium
ROOF BOUNDARY EXTRACTION USING MULTIPLE IMAGES
Author(s) -
Elaksher Ahmed F.,
Bethel James S.,
Mikhail Edward M.
Publication year - 2003
Publication title -
the photogrammetric record
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.638
H-Index - 51
eISSN - 1477-9730
pISSN - 0031-868X
DOI - 10.1111/0031-868x.t01-1-00003
Subject(s) - roof , photogrammetry , computer vision , artificial intelligence , computer science , epipolar geometry , boundary (topology) , matching (statistics) , pixel , consistency (knowledge bases) , position (finance) , aerial image , constraint (computer aided design) , building model , image (mathematics) , geography , mathematics , geometry , simulation , mathematical analysis , statistics , archaeology , finance , economics
Urban area building extraction is one of the most challenging problems in photogrammetry. Well‐extracted buildings are needed for a variety of applications, such as cartography, building GIS databases for cities, and urban planning. This paper presents a new technique to extract 3D building wire‐frames using a robust multi‐image line‐matching algorithm. Although one pair of images is adequate to find the 3D position of two visibly corresponding image features, it is not sufficient to solve the general building extraction problem due to obscured parts in the building. Four images are used in this research to extract the building wire‐frames. First the images are segmented into regions. Regions are then classified into roof regions and non‐roof regions based on their size, shape, and intensity values. The roof region boundary pixels are located and used to find the region perimeters. Region correspondence is solved in a pair‐wise mode over all images using the epipolar constraint, region size, region shape, and region intensity values. Image lines within the corresponding regions are matched over all images simultaneously by first creating a plane for each region line. Planes are then intersected simultaneously and geometric consistency is used to determine acceptance or rejection. Results with high overlap and sidelap aerial images are presented and evaluated. The results show the completeness and accuracy that this method can provide for extracting complex urban buildings. The average coordinate accuracy is about 0·8 m using 1:4000 scale aerial photographs scanned at 30 μ m. Six buildings were examined; the line detection rate is 98%.