Christopher O. Jaynes, Frank R. Stolle, Howard Schultz,
Robert T. Collins, Allen R. Hanson, and Edward M. Riseman,
"Three-Dimensional Grouping and Information Fusion for
Site Modeling from Aerial Images,"
Arpa Image Understanding Workshop,
Palm Springs, CA, February 1996, pp. 479-490.
The building reconstruction strategies that have been used in the UMass
ASCENDER system are reasonably effective, but are tuned to extract only
one generic building class with single-level, flat roofs bounded by rectlinear
polygonal shapes. Extensions to the system must be considered in order to
handle other common building types. Examples are multi-level flat roofs (or
single-level flat roofs containing significant substructures such as large air
conditioner units), peaked-roof buildings, juxtapositions of flat and peaked
roofs, curved-roof buildings such as Quonset huts or hangers, as well as
buildings with more complex roof structures containing gables, slanted
dormers or spires.
To achieve the desired goal of a more general and flexible building
extraction system in the ARPA/ORD RADIUS program, a significant
research effort is underway at UMass to explore alternative detection and
reconstruction strategies that combine a wider set of algorithms for
generating and fusing 2D and 3D information.
Click here for
full paper (2444323 bytes, compressed using gzip).