of DSM (Digital
en interested in a
DSM and DTM
and accuracy. In
1d the low cost of.
1 production cost
iatic approach for
n studied, which
ut imagery is the
tion costs do not
ed.
y and production
ly produced by
ery. Raster data
om the captured
een 1:15000 and
um. The pixel size
we an overlap of
rientation of the
plotter. Examples
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
2.3 Photogrammetric capture
Vectors can be of 3 types:
e Planimetry contours: 3D polygons
buildings, vegetation, water, bridges;
e Altimetry points and breaklines: 3D points and lines
describing the terrain and its orographic
characteristics in a very precise way (more numerous
when elevation varies); roads and railway connected
at intersections (crossroads, etc).
These 3D vectors are captured according to specific rules
depending on the type of object. For example, building contours
are captured at the gutter elevation. The minimal area is 25m?
and the minimal height above terrain is 2.5m. A building block
is divided into independent contours if the difference in height
describing
between them exceeds 2m. This division also applies for roof
superstructures, as long as area and height thresholds are valid.
Each polygon is associated to a 3D point captured at the highest
roof point (roof ridge or chimney). In addition, automatic
analysis guarantees that planimetry objects have a consistent
topology (closed contours, no intersection or self-intersection,
connected adjacent polygons). An example is given in figure 2.
Figure 2: Example of captured vectors (Kerlaz)
2.4 Raster data computation: DTM and DSM
The DTM is computed by triangulation from altimetry vectors,
including roads, railway, and water contours. Superimposing
aboveground elevations (buildings, vegetation and bridges) with
the DTM produces the DSM. Each polygon is associated to a
single elevation value corresponding to the highest point.
3. SEMI-AUTOMATIC PROCESS
31 Principle
| Vector capture |
Planimetry vectors
Altimetry vectors
Roads and railway
| Matching. algorithm A
Figure 3: Semi-automatic process
The semi-automatic process is based on an automatic matching
algorithm named AutoDEM that computes a DTM and a DSM
from two images and a few 3D vectors (see Figure 3). Input
vectors (altimetry and planimetry) are manually captured then
considered as external data by AutoDEM. Unlike the manual
process, raster data are computed using both vectors and source
images.
3.2 Matching algorithm AutoDEM
The matching algorithm consists of 4 steps (see Figure 4). It is
briefly described in the followings subsections and more details
can be found in (Baillard, 2003). It is characterized by the
intensive use of vector information at each stage of the process:
e Definition of an input elevation map (step 1),
eo Computation of local minimal and maximal z values
(step 1),
e Definition of an adaptive correlation window (step 2),
e Prior information for filtering raw DSM (step 3),
e Reference data for quality control (step 4).
Vector data
1. Pre-processing input data
yum = ——— 1 r…--------- a m zn in qu je
| Epipolar geometry ! | Vector analysis
Stereo pair
2. Image matching
Y
3. DSM analysis
DSM, DTM
4. Self-evaluation
Figure 4: Automatic computation of DTM / DSM with
AutoDEM
3.2.1 Pre-processing input data (images and vectors)
Each image pair is resampled into epipolar geometry. The
vectors are analysed in order to produce a set of reference
vectors (used for quality control) and a set of “input maps”:
input elevation map, building maps, minimal and maximal
elevation maps. Finally source images and input maps are sub-
sampled in order to create image pyramids.
3.2.2 Image matching: computation of a “raw” DSM
The matching algorithm uses dynamic programming within a
multi-resolution scheme, which is particularly appropriate to
dense urban scenes (Baillard, 2003). Input maps are taken into
account as follows:
e The input elevation map constrains the research for an
optimal path by defining input matched points,
e The minimal and maximal elevation maps define
allowed and forbidden areas for the path,
e The building maps are used to weigh the correlation
score between 2 pixels.
3.23 Analysis and filtering of “raw” DSM
A set of reliable ground points is first selected from the DSM
by combining various criteria and filtering methods: Top hat
morphological filtering, small height relative to neighbours,
