According
to the American Society of Civil Engineers (ASCE), more
than 26% of the nation’s bridges in 2009 were classified as either
structurally deficient or functionally obsolete. Two years later,
structurally deficient or functionally obsolete bridges still
made up close to 24% of the nation's total bridge infrastructure.
A report by the Federal Highway Administration (FHWA) indicates that,
given more time and funding to complete bridge inspections, the use
of non-destructive evaluation (NDE) methods would increase among
state and county transportation agencies (Highway
Bridge Inspection: State-of-the-Practice Survey,
2001). NDE promises a way to enhance the allocation of funding by
improving the information these decisions are based on and by
improving the assessment of existing bridge conditions and through
increased safety of inspection crews, reducing traffic disruption,
and increasing the frequency, objectivity, and accuracy of bridge
condition assessment.
As
part of research funded by USDOT-RITA, the 3D Optical Bridge
Evaluation System (3DOBS) was developed to quickly assess the
condition of bridges while minimizing traffic disruptions and
limiting inspection crews' exposure to traffic. The system is
composed of a Digital Single Lens Reflex (DSLR) camera mounted on a
truck, close range photogrammetry software (Agisoft PhotoScan Pro)
and an automated spall detection algorithm. For close range
photogrammetry to be achieved, the photos need to be collected with
at least 60% overlap. Early testing of the photogrammetry software
showed that collecting imagery with greater overlap produced better
results.
Prior
to the collection of photos, the bridges had to be marked with
reference points for Agisoft to set up a coordinate system and to
create a DEM. These reference points were marked duct tape that was
placed on the bridge deck in a grid pattern. The tape was placed at
four foot intervals in the transverse direction and at ten foot
intervals in the longitudinal direction. Carrier phase GPS points
were collected with a Trimble GPS (with an accuracy of <1m) at
each of the four corners of the bridge deck and at various other
points on the deck to be able to correctly spatially reference our
data.
For
the collection of the photos, a standard consumer grade Nikon D5000
DSLR with a resolution of 12.3 megapixels (MP) and a 27 mm focal
length lens were used. In order to capture a full lane in one pass
the camera needed to be mounted 9 ft above the bridge deck. In order
to achieve this height, a wooden vehicle mount was constructed to fit
into the bed of a standard pickup truck. During field collections a
control board was programmed to trigger the camera shutter at a rate
of one image per second. With the camera mounted, the truck was
driven across the bridge deck at a speed of about 2 mph. This speed
ensured that images were captured with the required 60% overlap
between the photos.
After
the photos were collected they were processed in Agisoft PhotoScan
Pro. This process was mostly automated as the software aligned the
photos and generated a 3D model without any user input. After the
model was generated it was necessary to manually add "KeyPoints"
to mark the location of each one of the duct tape markers with the
latitude and longitude coordinates. This allowed PhotoScan to set up
a coordinate system and to accurately reference the model and create
a Digital Elevation Model (DEM). The DEMs that were generated have a
resolution of 5 mm in the x, y directions and a z resolution of 2 mm.
The
spall detection algorithm was written in Python programming language
and uses ArcPy to interface with ArcGIS and utilize some of ESRI’s
available geospatial tools. The tool used to detect spalls is called
Focal Statistics analyzes each cell in the raster and calculates
statistics based on a specified neighborhood of cells around it.
Additional functionality was added so that the user can remove bridge
joints by creating a shapefile that defines the bridge joints. Spalls
can also be identified based upon their area. This feature allows for
the detected spalls to correspond to minimum size definitions and
allow for the removal of small artifacts in the DEM. The data
processing is automated, and only requires the user to set the
working directory, file names for the DEM and bridge joint shapefile,
focal statistics sensitivity and the minimum spall size.
Enhancements
are currently being made to this system through a project sponsored
by the Michigan Department of Transportation (MDOT). These
enhancements will include improving the camera so that the system
would be able to operate at near highway speeds (40 mph) and the
construction of a more sturdy vehicle mount. The spall detection
algorithm will also be improved. The current version of the algorithm
simply looks for a change in the elevation and therefore it also
detects the edges of patches on the bridge and reports them as
spalls. This will be changed so that the algorithm will only detect
negative changes or those that represent spalls on the bridge deck.
This
work is supported as part of a larger program (Bridge Condition
Assessment Using Remote Sensors) sponsored by the Commercial Remote
Sensing and Spatial Information program of the Research and
Innovative Technology Administration (RITA), U.S. Department of
Transportation (USDOT), Cooperative Agreement # DTOS59-10-H-00001,
with additional support provided by the Michigan Department of
Transportation, the Michigan Tech Transportation Institute, the
Michigan Tech Research Institute, and the Center for Automotive
Research. The views, opinions, findings, and conclusions reflected in
this paper are the responsibility of the authors only and do not
represent the official policy or position of the USDOT/RITA, or any
state or other entity. Further information regarding remote sensing
technologies and the decision support system for bridge condition
assessment and about this project can be found at
<http://www.mtri.org/bridgecondition>.
