Shoichi Horiguchi
roof faces in a robust manner and the reconstruct the topological relationships between the faces. We presented an
approach for the reconstruction of prismatic building models with flat roofs using surface analysis and Hough
transformation (Horiguchi, S. et al, 1999). However, urban models should not include only buildings but also roads,
blocks, sidewalks and roadside trees and so on. The first two roads and blocks, very important for producing useful
urban simulations.
1.3 Our Approach
; START
This paper describes a new approach to reconstruct road, (starr)
intersection and block models: we use the road network on the
digital 2D map. In the road network, a node means intersection, a
link means a road. A block is extracted as an area enclosed by roads
and intersections. When reconstructing road, intersection, and block
models there are three problems. The first is the separation of road
and block areas. The second is the determination of road and block
model parameters. The third is reconstruction of optimal model, that
is, both model size and the error must be minimized. These three
problems are settled by the process flow shown in Figure 1.
First, we separate road and block areas by matching the edge points
of buildings extracted from DEM on the digital 2D map.
Second, we extract the surface model primitives of roads, |
Matching DEM with Map
Y
Extraction Surface model
Y
Reconstructing Surface model |
intersections and block.
Third, we assume that the intersection surface models as flat, and
reconstruct the road surface models using the Minimum Description
Length (MDL) principle. A road surface model reconstructed by
this principle has both minimum model degree and the model error.
We reconstruct block surface models as polygons delineated by the
vertices of the adjoining intersection models and road models.
Fourth, we combine surface and building models. Building models
are reconstructed using the technique presented in (Horiguchi et al.,
1999),
Finally, we project the realistic texture acquired from aerial video
images onto the models of buildings, roads, intersections, and block
surfaces. We proposed a technique that yields exact texture mapping
by comparing the polygons of 3D models to the texture shapes
using MDL principle
Y
Combining Surface model
with Building model
Y
Texture Mapping
within aerial video images (Horiguchi et al., 1999). We try to ensure END
robust, high quality texture projection. In order to determine the
capture system parameters, video camera motion, position, and Figure 1. Block Diagram
orientation are determined using a Differential - Global Positioning
System.
In a trial, 3D urban models of roads, intersections, block surface models and building models are recovered by our new
approach.
2 RECONSTRUCTING OF ROAD AND BLOCK SURFACE MODEL
2.1 Matching DEM with Digital 2D map
We match DEM to a digital 2D map using the building shape information in the digital 2D map in order to extract road,
intersection and block areas from DEM. Although DEM and the digital 2D map have grid values based on common
coordinate system together, it is not easy to match them. When matching DEM to a digital 2D map by using only grid
values, the displacement errors become very large because the DEM is inaccurate in the horizontal plane. Therefore,
this paper describes an approach that minimizes the errors.
We previously proposed a technique to divide DEM data into three surface types: Flat type, Slope type, and Boundary
type by using Gaussian and Mean curvature (Horiguchi et al., 1999). In this case, the point sets of Boundary type
represent the contours of buildings, and corresponds building shapes on the digital 2D map. Therefore, our approach
minimizes the distance between points of Boundary type and lines of building shape by offsetting the DEM alignment
on the horizontal plane.
Figure 2 shows an example of DEM - digital 2D map matching. In Figure 2, the left image shows the Boundary point
414 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
