Automating Near Real-Time, Post-Hazard Detection
of Crack Damage to Critical Infrastructure
Eugene A. Schweizer, Douglas A. Stow, and Lloyd L. Coulter
Abstract
This study seeks to determine the most effective operational
approach for detection and delineation of fine-scale crack
damage to critical infrastructure using rapid, semi-automated
image analysis methods. The effectiveness of these methods is
tested using both single-date (post-event only) imagery and bi-
temporal (pre- and post-event) image pairs to validate benefits
associated with the collection of baseline imagery and corre-
sponding camera station locations needed to employ bi-tem-
poral, change detection based on the repeat station imaging
(
RSI
) approach. This study is the first to compare a variety of
image processing methods of detecting and delineating cracks
using bi-temporal, aerial images of complex scenes as input.
Methods employing a bi-temporal, change detection approach
outperform techniques relying on single-date imagery only.
With a producer’s accuracy of 77.9 percent and user’s ac-
curacy of 69.1 percent, a pixel-based image difference model
using kernel based spatial filters is the most successful and
is suitable for automated, near real-time implementation.
Introduction
Across the world, both natural and anthropogenic disasters
have killed more than 900,000 people and caused over $1.5
trillion in damage since 2005 (The International Disaster
Database,
CRED
, 2015). While most disasters that cause such
massive losses cannot be accurately predicted, the response
to them can be improved. After a disaster, there are several
phases that affected communities go through, called the emer-
gency response cycle. This cycle begins with the response
phase, which includes rescue, relief, and recovery, and later
moves into the mitigation phase, which includes reconstruc-
tion and the improvement of preparedness for future disasters
(Cutter, 2006). Using the increasing power of geographic infor-
mation science and remote sensing technologies to inform our
understanding of extreme events such as natural disasters, the
effectiveness of actions taken in both the response and mitiga-
tion phases of the emergency response cycle can be improved.
Remote sensing has become an increasingly useful tool for
improving the situational awareness of first responders and
emergency management personnel during and immediately
following extreme events such as natural disasters. The abil-
ity of remote sensing technologies to derive information from
imagery depicting damaged areas that may not be accessible
to other methods makes it a uniquely valuable tool (Cutter,
2006). Also, no other survey method provides such a spatially
comprehensive, synoptic view of potentially damaged areas
as does remote sensing (Tralli
et al.
, 2005). To maximize the
unique ability of remote sensing derived products to increase
awareness and improve response after a disaster event, a rap-
id, automatic method of assessing post-hazard damage needs
to be developed. Through this research, we have designed
and tested techniques for the semi-automatic assessment of
post-hazard crack damage to critical transportation infrastruc-
ture that can be applied in near real-time.
Providing the specific information required by the user is
the primary function of remote sensing (Stow
et al.
, 2015).
While studies such as Phinn
et al.
(2003) have detailed a
hierarchical structure used to tailor the collection of the most
appropriate data to the needs of the end user, the ability
to rapidly process and disseminate remotely sensed image
products to decision makers has not been widely addressed
(Lippitt
et al.
, 2014). To this end, Lippitt
et al.
(2014) have
expanded on the model of Strahler
et al.
(1986), to examine
the variable of timeliness as a key measure of utility. When
the positive impact of information on the decision-making
process decreases with time, each link in the “image chain”
(Schott, 2007) must be evaluated to ensure no time is being
wasted (Lippitt
et al.
, 2014).
Sensitivity to timeliness is particularly important during
the response phase to extreme events such as natural disas-
ters, as rapid action is needed to minimize the loss of life and
destruction of property. With the growing availability of very
high spatial resolution digital imagery, as well as improved
methods for rapid data transfer, the ability of remote sensing
to inform decisions in near real-time is increasing (Coulter
et
al.
, 2011; Talbot and Talbot, 2013; Visser and Dawood, 2004).
With these improvements in data collection and distribution
being realized, there is now a need for an improved method
of identifying specific manifestations of damage to critical in-
frastructure that leverages these advancements. The ability to
detect and identify, in near real-time, specific damage features
such as cracks in road and bridge surfaces will improve the re-
sponse to catastrophic events by rapidly providing specific, ac-
tionable information on the condition of critical infrastructure
to those emergency managers coordinating response efforts.
The goal of this research is to develop and test different
single- and bi-temporal image analysis methods capable of
rapidly detecting, and potentially identifying manifestations of
damage, specifically cracks in critical infrastructure. The prod-
ucts generated could be used by image analysts to establish
which structures exhibit evidence of damage, assisting emer-
gency response personnel in determining which structures to
prioritize for manual inspection, or if it is necessary to divert
emergency response efforts away from such infrastructure. The
products created in this study do not seek to replace on-site
inspections to determine the safety of critical infrastructure
after a hazard event, but rather provide a rapid preliminary as-
sessment. If immediate emergency response decisions (e.g., di-
verting ambulances or evacuees away from damaged bridges)
Eugene A. Schweizer is with San Diego State University, 1762
Reed Ave., San Diego, CA 92109 (
).
Douglas A. Stow and Lloyd L. Coulter are with San Diego
State University, Storm Hall 307B Department of Geography,
San Diego State University, San Diego, CA 92182-4493.
Photogrammetric Engineering & Remote Sensing
Vol. 84, No. 2, February 2018, pp. 75–86.
0099-1112/17/75–86
© 2018 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.84.2.75
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2018
75
