Methodology
PCA
and multiple linear least square regression models were
used to model the relationship between the dependent vari-
able of overall distress rate (
ODR
) and explanatory variables
extracted from the spectral response of the
HSR
multispectral
digital aerial photography. The ultimate goal is to be able to
predict the
ODR
for roadway segments for which
ODR
ground
reference values are unavailable.
Data Acquisition and Preparation
The study area for this research encompasses six counties in
northern New Mexico, including Bernalillo, Cibola, Sandoval,
Santa Fe, Torrance, and Valencia. These counties are located
around the City of Albuquerque and are covered by all of the
existing
HSR
multispectral digital aerial photographs with var-
ious spatial resolutions obtained from 2004 to 2012. Within
the study area, 50 data collections sites were identified for use
in this study. Each data collection site covers the rightmost
lane with a length of one tenth of a mile. These sites were
selected because they belong to a set of pavement sections
regularly evaluated as part of the New Mexico Department of
Transportation (
NMDOT
) pavement evaluation program
.
The road segments of these 50 study sites were visually
evaluated in reference to the available
HSR
multispectral digi-
tal aerial photographs to ensure they were covered by aerial
photographs, and there are no large obstacles (e.g., bridges
and overpasses) above them. A geographic information sys-
tem (
GIS
) database provides the roadway number, milepost
number, and direction for each study site.
Reference pavement surface condition data for the study
sites were acquired from records of manual pavement evalu-
ations conducted for
NMDOT
during the summer of 2009
(Cordova
et al.,
2009). All of the study sites were constrained
to flexible, asphalt pavements. For flexible pavements, the
NMDOT
evaluates severity and extent of the following seven
distresses on a scale of 0 to 3 (0 = Not Present, 1 = Low, 2
= Medium, and 3 = High): (1) Raveling and Weathering; (2)
Bleeding; (3) Longitudinal Cracking;(4) Transverse Cracking;
(5) Alligator Cracking; (6) Edge Cracking; and (7) Patching.
It should be noted that the listed distresses are all horizontal
distress, and they do not reflect distresses in elevation such as
rutting and shoving. This makes the use of
HSR
multispectral
digital aerial photographs to detect pavement surface overall
distress rate (
ODR
) possible since elevation information cannot
be found in a typical aerial photograph (the exception being
stereoscopic aerial photographs)
.
Each study site’s
ODR
can be calculated based on the fol-
lowing equation:
ODR
=
∑
7
i
=1
(
α
i
×
β
i
×
γ
i
)
(1)
where
i
represents each of the seven distresses,
α
denotes the
severity rating,
β
denotes the extent rating, and
γ
denotes the
weighting factor. The weighting factors for the distresses have
been provided by
NMDOT
and are 3, 2, 12, 12, 25, 3, and 2,
respectively, for each of the seven distresses. The calculated
ODR
for each of the 50 study sites ranges from 0 to 477. The
lower the
ODR
value, the better the pavement condition. The
maximum possible value is 504.
ODR
can be easily converted
to Present Serviceability Index (
PSI
), which is broadly used by
various transportation agencies (Bandini
et al.,
2012). Differ-
ent agencies can develop and establish their own models to
infer the overall pavement surface conditions, no matter what
particular metric they are using
.
One set of archived and readily available ortho-corrected
HSR
multispectral digital aerial photographs with a spatial
resolution of 6-inches were obtained from the Earth Data
Analysis Center (
EDAC
) at the University of New Mexico. The
aerial photographs are natural color digital aerial photography
that records energy in the region from 0.4-micrometer to
0.7-micrometer range, and they have three spectral bands
which include visible blue (0.4 to 0.5 micrometers), visible
green (0.5 to 0.6 micrometers) and visible red (0.6 to 0.7 mi-
crometers) (Jensen, 2007). These images are in 8-bit data for-
mat and are the actual digital numbers recorded. In addition,
these aerial photographs are routinely (every the other year)
collected with the Zeiss/Intergraph Digital Mapping Camera
(
DMC
) System by the Mid-Region Council of Governments
(
MRCOG
) contracted to Bohannan-Huston, Inc.
The aerial photographs were taken in March through April
2010 and were matched with the manually-collected pave-
ment condition data collected in May through August 2009.
This was the closest time match between the aerial photo-
graphs and the pavement condition data available. According
to the Federal Highway Administration, it is approximately
a 15-year process for pavement surface condition to drop 50
percent in quality (Lenz, 2011). Because the time elapsed
between the pavement condition data collection and aerial
photographs collection was less than a year (approximately
six months), we assume that no significant change occurred at
the study site.
Image Processing
Image Aggregation
Data on actual pavement surface conditions were collected on
short sections (0.1-mile [161-meters]) of pavement located at
specific mileposts. In order to identify the evaluation zone of
each study site on the aerial photographs, a buffer of 0.1-mile
was created around each individual study site’s milepost in
the aerial photographs. After creating the buffers, the evalua-
tion zone of several study sites could not be completely cov-
ered by a single 6-inch image, because the aerial photographs
were divided into tiles to reduce the storage size. In this case,
two or more photographs were needed. When multiple pho-
tographs were used for a single milepost, it was necessary to
create a mosaic of the aerial photographs. These images were
mosaicked based on standard overlay-based algorithm and
average blending mode.
Evaluation Polygon Creation
Pavement surface conditions are only evaluated within a
portion of the roadway. According to the protocol for manual
evaluation employed by
NMDOT
(Cordova
et al.,
2009), pave-
ment surface distress data were collected only in the right-
most driving lane and never in passing lanes, turning lanes,
or on the shoulder. For two-lane roadways (one driving lane
in each direction), data were collected only in the positive di-
rection (north and/or east) from a given milepost to a distance
of 0.1-mile. For multi-lane roadways (two or more driving
lanes in each direction), data were collected in both the posi-
tive and negative directions (north-south and/or east-west)
at a given milepost. In the positive direction the pavement
evaluation was conducted from a given milepost to a distance
of 0.1-mile, while in the negative direction the evaluation was
conducted from 0.1-mile prior to the given milepost. This
ensured that the pavement sections evaluated at the given
milepost were parallel and aligned to each other.
To ensure alignment between the data collection zones,
polygons were created to represent the highway zones used
in manual evaluation, and from here on referred to as “evalu-
ation polygons.” These evaluation polygons were created by
heads-up digitizing over the 6-inch resolution aerial photo-
graphs following the protocol mentioned above. It should be
noted that these manually created polygons only cover the
pavement surface, and the polygon creation process does not
involve any removal of the non-road surface elements (e.g.,
vegetation). Therefore, there is no classification involved in
the analysis. In addition, there are thousands of pixel cells in
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
September 2015
711
