In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009
Figure 5. Edge detection & matching in an urban environment
on Ikonos imagery.
At a final stage a least-squares matching method, called
modified multi-photo geometrically constrained matching
algorithm, is performed using all matched points as
approximations to detect mismatches and to further refine
matching results. The MPGC algorithm combines the matched
points with geometrical constraints, derived from multi-image
ray intersection conditions and knowledge about the image
orientation (Baltsavias, 1991). A Least Squares B-Spline
Snakes is used to refine the matched edges. For more details on
the matching strategy we can refer to (Zhang & Gruen, 2006).
During image matching, calculation of the position and height
of each point or line is treated independently. To create a
connected surface, the discrete measurements are interpolated.
The resulting surface model is processed at a grid size of 3
meters. The chosen resolution leads to the best equilibrium
between detail and reduction of noise. As illustrated in figure 6
& 7, the shape of big buildings and free-standing buildings is
modelled well, while in the very dense urban area small
buildings are merged into building blocks.
Figure 6. Map view on extract of the 3m colour-coded DSM.
Figure 7. Perspective view on extract of 3m color-coded DSM.
The surface model represents Istanbul's historic peninsula.
4.6 Ortho-generation
During ortho-generation phase the sensor geometry of the
images, characterized by a parallel projection in along-track
direction and perspective projection in across-track direction,
can be transformed to map geometry based on the developed
surface model. The surface model represents each pixel in its
correct geometric position. Back-projection from the DSM to
the image supplies the grey value or texture for the pixel. In
case of an occluded pixel on the master image, texture
information is extracted from a slave image or neighbourhood
pixels in case of occlusion on all images. A ground sample
distance of 1 m or 1 pixel is chosen for the ortho-image.
Figure 8. Extract of 3m surface model, draped with
panchromatic ortho-image for photorealistic visualization. The
surface model represents Istanbul’s historic peninsula.