Actual Building
/ and result of
building refinement
User delineation is
done very rapidly
N
Figure 1. Building refinement
6.2 Building Extraction Tests
Figure 1 shows an example of the building refinement tool. This
tool takes "seed" points from the user that are dropped within a
few pixels of the actual building corners. From this quick
drawing, the building is "refined" in a few seconds using edge
finding, Z finding, and squaring processes. The result is
consistently a precisely collected building in less than half the
typical time. This was demonstrated by having a reasonably
experienced stereo compiler extract building tops for many
hours using conventional photogrammetric extraction and
comparing this to the semi-automated method. These tests were
performed using 1:11,000 scale images in very rolling terrain.
The typical buildings were houses of differing sizes and
orientations. Some buildings had trees around them and in
some cases the trees were overhanging portions of the buildings.
The results were quite startling in that the user was able to
extract more than twice as many buildings in the same amount
of time. Even more startling was the user opinion of fatigue
reduction. This user felt that the fatigue reduction was very
substantial when compared to conventional extraction. This
might be expected when one considers that the majority of the
labor in vector extraction is in placing the floating mark
precisely in three dimensions. The last few pixels of placement
are the most time consuming and fatiguing. When using semi-
automation, the user is permitted to be fast and sloppy when
placing the cursor in all three dimensions. As long as the user is
within the refinement algorithms search distance (such as three
pixels) the speed and robustness of the tools are very good. Of
course more tests are necessary to better quantify the average
productivity improvement.
6.3 Automated tools in the product or in testing
These types of refinement tools are now being brought into the
commercial product. Several tools are now available for linear
and polygon feature collection. The following are some example
refinement tools:
1. Building refinement with and without Z finding
2. Homogenous area (lakes, ponds, etc.)
3. Vegetation/Tree regions
4. Trails and centerlines
5. Road and boundary edges with and without
thinning/filtering
254
user measures ct uu
ee RT
pts. neartrail _y
«+ Resulting
delineation via
semi-automation
Figure 2. From a few seed points, a detailed feature can be
drawn
Figure 2 shows how a detailed feature such as a trail can be
extracted in detail using just a few seed points. Approximate
points can be mixed with precise user defined points. The semi-
automatic process will then precisely derive the feature based
on image and logical processing.
In Figure 3, a man-made road edge is depicted and a small
number of seed points are measured near the road edge. The
image processing and logical operators will find the transition
points and return a thinned and smoothed delineation.
0
Seed points do not need
to be measured precisely [—— —,
on the feature
tenez
Figure 3. Road collection and refinement
6.4 Context Sensitive
One of the challenges in developing these tools is to keep the
user interface streamlined. The "tweaking" of parameters and
selection of different tools must be minimized or the user
becomes overburdened just setting up the desired tool. One
method that we are using to help in this area is the concept of a
"strategy" for each tool type. The strategies allow several
permutations of algorithms but they can be named in a user
intuitive way. This permits easy switching from one tool type to
another. In addition, the user can setup the default collection
mode on a per "class" or "feature type" basis. This permits the
software automatically to set the desired collection mode based
on the context of the feature class or type. This eliminates the
step of changing collection modes during typical extraction. For
example, the default collection mode for houses and industrial
buildings might be "semi-automatic", while the default
collection for churches might be "manual" (assuming churches
are too complex for semi-automation). Another example might
be "semi-automatic center-line" for trails and "semi-automatic
spline edge" for streets or highways.
Overall, this semi-automated approach is proving to be user
friendly and offers genuine enhancements to productivity for
vector extraction. Much work will continue in this area and
timeline improvements in the range of two to four times appear
to be plausible.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
It is
pho
dev
fair
witl
bee!
con
pho
rem
