STATEMENT
OF THE PROBLEM
A major stumbling block to the integration of remotely sensed data
into existing GIS data base structures is the issue of positional
accuracy of the existing line-work within the vector database. This
inaccuracy manifests itself when overlain to more positional consistent
imagery data. Figure 1 shows a typical situation occurring between
the existing GIS data layer (parcels) and the imagery base.
Figure 1: Vector overlay on IKONOS
PRESENT
SITUATION
In this case the parcel map had a variable accuracy of up to 40
ft plus or minus once the various tiles were combined. This is the
result of data being built by hand historically remaining un-edgematched
between tiles within a mylar mapping system. The investment to convert
this (the only base map widely used) was made and the sheets were
scanned and vectorized by the private sector, which very accurately
reproduced the inherent errors of this mapping approach. With the
incorporation of GPS and the associated problems of edgematching
the tiles into a seamless database the consortium was stymied. This
lead to the development of an image based reference for these data
layers from the existing DOQQs (1995 vintage) (Figure 2).
Figure 2: 1995 Digital Orthophotography
POTENTIAL BENEFITS
Through the creation of a single base image map the respective users
and developers of the numerous data sets being created and compiled
within the county, city, and private sector, could be spatially
integrated and the investments being made across within the region
maximized.
TECHNICAL APPROACH
TAKEN
A process was developed that assisted the county in creating a seamless
database that could be used by any local government or public works
group as an accurate and reliable base map from which all other
vector data could be migrated while maintaining the integrity of
the relative-positional accuracy of the vector line-work. The process
involved the original acquisition of the DOQQ data from the Missouri
Spatial Data Information Service (MSDIS). The DOQQ data was delivered
to MSDIS on CD-ROM. The DOQQ data for Boone County came in the form
of the original unprocessed DOQQ data. These were the DOQQs that
were used for the base map portion of the BOCOMO project. Processed
DOQQs were also delivered to MSDIS on CD-ROM. These processed DOQQs
had been linear stretched and appeared visually unappealing when
they were placed together in a seamless database. The differences
in the contrast from one DOQQ to the next DOQQ were unappealing.
Therefore, it was decided to use the unprocessed DOQQs for this
project.\
The original DOQQs had to be processed using an AML created to
read the USGS data format and convert the DOQQs into a readable
image file. The AML created Band Interleaved by Pixel (BIP) formats.
The next step involved converting the BIP DOQQs into Arc/Info grids.
This was done using the imagegrid command in ARC. Imagegrid converts
images into grids. These grids were then projected from their original
format (Projection UTM, Zone 15, Datum NAD83, Units Meters) to their
new projection (Projection Stateplane, Zone 4426, Datum NAD83, Units
Feet). This was done because Boone County currently uses the Stateplane
projection for all of its spatial data. The projection gives different
options as to how to perform the resampling. Cubic convolution was
used as the resampling procedure because it was felt that cubic
convolution maintains the spatial integrity of the data, although
it does tend to interfere with the spectral integrity. It was felt
that maintaining the spatial integrity of the DOQQs was more important
than maintaining the spectral integrity of the DOQQs because the
DOQQs were ultimately going to be used for the county wide base
map. Therefore, cubic convolution, as opposed to nearest neighbor,
was selected as the resampling process used when projecting the
grids. Once the grids were projected to Stateplane, they could then
be converted back into images so that further processing could be
performed on them in an image processing software. The gridimage
command was used to convert the grids into an image format that
could be read in ENVI image processing software. The image file
selected was the TIFF file format. The TIFFs were then brought into
ENVI.
The TIFF DOQQ image files were then histogram matched before any
warping or other processing was done. The histogram matching was
done to make the DOQQs more visually appealing. Histogram matching
was done in order to make the brightness distribution of the two
images as close as possible. All of the DOQQs were histogram matched
to the same originating DOQQ. This DOQQ was Brownsne.tif. The northeastern
quadrant of Browns was chosen as the matching DOQQ for the rest
of the DOQQs because it contained a distribution of rural to urban
land that was proportionate to the county as a whole. Once all the
DOQQs were histogram matched an evaluation began that used the centerlines
that were collected by the Boone County Public Works Department
as a source of ground truth. The centerlines were created from their
GPS kinematic data collection that was done for the whole county.
These lines were not the exact centerlines because the vehicle used
in the collection process often had to drive down one side of the
road or the other. Nevertheless, the lines should when overlain
on images fall within the road encasement. The evaluation involved
overlaying the centerlines on various DOQQs throughout the county
to see if they fell within the encasements of the roads. The majority
of the areas that were visually inspected showed that the centerlines
did in fact fall within the encasement of the roads. Nevertheless,
a ground team was activated in order to collect image based ground
control points for various portions of the county. Over 200 ground
control points were collected for the majority of the central to
southern portion of the county. These ground control points were
used to try and warp the individual DOQQs to see if warping them
would produce more accurate x and y results. The results of warping
the histogram matched DOQQs were encouraging when looking at the
individual DOQQs, however, when trying to mosaic the warped DOQQs
they would not line up properly along the edges of the DOQQs. A
further evaluation of the ground control points along with the centerlines
showed that there was an error of approximately 4 to 6 feet, between
what was considered truth (the GCPs and the centerlines) and the
DOQQs. The county and city requirements were that the DOQQs be positionally
accurate within 10 feet of ground truth. Because the DOQQs were
within the range defined by the county and city, it was decided
not to try and make the DOQQs any more positionally accurate (Figure
3).
Figure 3: Unwarped DOQQs with centerlines overlain
The next step involved in the DOQQ base map was the mosaicking
or all the DOQQs into one seamless database. The individual DOQQs
were first mosaicked together into their respective quads. Then
each of the quads was put together in tiles running north to south.
Once these tiles were put together, they were then mosaicked together
into one database. In all, there were around sixty individual DOQQs,
which were mosaicked into 15 quads, which were mosaicked into 5
tiles, and finally which were mosaicked into the DOQQ base map database.
Each time a mosaic was done, feathering was performed to help smooth
out the seams along the edge of the DOQQs. A one hundred-meter feather
was applied to each mosaic performed during the building up process
of the base map. Figures 4 and 5 show a comparison of a non-feathered,
non-histogram matched mosaic section, with an example of a feathered
mosaic section.
Figure 4: Non-feathered, non-histogram matched DOQQ mosaic
Figure 5: Feathered and histogram matched DOQQ mosaic
Once the DOQQ base map was completed, the IKONOS (April 2000)
imagery was integrated into the DOQQ database to give a more current
and up-to-date idea of what was occurring in the county. The IKONOS
database covered roughly one half of the county. Its geographical
extent ranged from the north central portion of the county down
through the southern portion of the county. The lowest precision
IKONOS was acquired from Space Imaging. Two separate swaths were
purchased, an eastern swath and a western swath (Figure 6).
Figure 6: IKONOS coverage for Boone County
The IKONOS imagery had to be co-registered to the DOQQ base map
database. Before co-registration began, the IKONOS had to be projected
from its original UTM projection to Stateplane, as was the case
for the DOQQs. Co-registration was labor intensive and tedious.
Roughly 1600 image-to-image registration points were selected to
register the IKONOS to the DOQQ database. So many points had to
be used because the integrity of the spatial accuracy was at stake
when registering to the DOQQ database. As well, the IKONOS imagery
was not ortho-rectified so we were attempting to accomplish both
the registration and ‘orthofication’ of the IKONOS imagery
with this process. When fewer points were tried, the IKONOS would
lose some of its spatial accuracy. Using more points increased the
co-registration accuracy of the IKONOS. It is important to note
that the maximum accuracy of the IKONOS co-registration would only
be as good as the base accuracy of the DOQQs. The resulting IKONOS
accuracy was roughly similar to the DOQQs (approximately 5.5 feet).
Once the IKONOS was registered to the DOQQs, it was possible to
integrate the IKONOS with the DOQQ to get a seamless database with
the IKONOS coverage overtop of the DOQQ data (Figure 7). Once these
DOQQ database and the IKONOS/DOQQ database were completed, each
were compressed using the commercial product MrSID in order for
file transfer and distribution to be made easier.
Figure 7: IKONOS (right) and DOQQ (left) Mosaic
with centerlines overlain
PRODUCTS CREATED
The deliverables for this segment of the BOCOMO project was a countywide
image base with a user defined spatial accuracy of plus or minus
10 feet. The deliverables included:
1. Countywide 1995 DOQQ seamless mosaic (RMSE 5.1 feet)
2. Countywide 1995 DOQQ seamless mosaic with IKONOS data integrated
where coverage was available (RMSE 5.5 feet)
3. Vector migration protocols for conversion of spatially inaccurate
vector data to the image base.
ANTICIPATED IMPLEMENTATION
PROBLEMS IN LOCAL GOVERNMENT
The users will still not be able to see all the features that they
would like to. A procedure for building tools to allow for inventory
of directly visible objects (transmission towers, bridges, maintenance
buildings, etc.) as well as referential inventory of objects too
small for the IKONOS imagery to delineate (fire hydrants, utility
poles, storm water inlets, etc.) is needed and will be sought in
Synergy Phase II. Work will also need to continue with the users
to define decision rules whereby the conversion process for the
migration of vector files to the image base can be automated to
the extent possible. This is critical work as it is this migration
and spatial referencing capacity which will allow the legacy systems
and data bases to be integrated with the remote sensing data available
today. The tool kits and interfaces to allow for this conversion
process will be pursued within in Phase II. As well, training and
education in the use and utility of these data is needed to ensure
proper implementation and integration within the local government
applications.
ADDITIONAL WORK
1. Tool kit development for inventory of features from the base
map
2. Decision trees for the automation of vector migration tools
3. Feature extraction from the image base for new features (roads,
building footprints, etc.)
4. Continuing study of accuracy assessment and cost/benefit analyses
based on various methods of building and maintaining an image base
map.
PROJECT PARTICIPANTS
Timothy L. Haithcoat, Dan Daugherty, Jim Dunajcik, and Derek Smith,
Geographic Resources Center (GRC), Department of Geography, University
of Missouri under the auspices of ICREST. User clients included:
Ross Short, City of Columbia and Boone County GIS Coordinator; David
Storvick, City of Columbia, Engineering Department; J.R. Richardson,
Boone Electric Cooperative;
|
