Computer Vision and Pattern Recognition Technologies for Construction - Mobile and Pervasive Computing in Construction - page 198

10.3.3 Concrete surface discoloration
Another defect that reduces the quality of concrete surfaces is discoloration, which
is defined as departure of color from the normal or desired concrete surface color.
Discoloration on concrete surfaces has no specific shape, no sharp boundaries, and
no certain textures. Moreover, according to its gray-scale histogram, no obvious
threshold value can be selected to differentiate discoloration defects from the
normal concrete areas. Facing this problem, Zhu and Brilakis (2010b) present an
approach that contains two steps. Firstly, the degree of discoloration defects on a
concrete surface is globally measured by calculating the standard deviation of color
values of a concrete surface image. The larger/smaller the standard deviation, the
more/less the color of the inspected concrete surface departs from that required. If
the standard deviation is larger than a given threshold, it indicates that discolor-
ation exists on a concrete surface. Secondly, image segmentation and region
comparison are used to differentiate discoloration regions from normal concrete
surfaces. Validation results for this automated method are almost the same as
manual inspection results performed by experienced inspectors.
10.3.4 Rebar exposure
In post-earthquake structural safety evaluations the detection of exposed reinfor-
cing steel is of utmost significance. German et al . (2011) present a method of
automatically detecting exposed reinforcement in concrete columns for the pur-
pose of advancing current practices. Under their method, the binary image of the
reinforcing area is firstly separated using an adaptive thresholding technique. Next,
the ribbed regions of the reinforcement are detected by way of binary template
matching. According to the resulting image, and common dimensions of the
reinforcement in relation to concrete columns, both horizontal and vertical
profiling is performed, resulting in the combined binary image disclosing only
the regions containing rebar. Based on test results, this method can correctly detect
83% of exposed reinforcement.
10.3.5 Steel surface
The most common defect on a steel surface is corrosion. Cheng and Chang (2002)
use artificial neural networks to recognize rust defect areas in gray-scale images of
steel bridge coating. In their method, image thresholding is the key technique to
producing a binary image that separates rust regions from other areas. In this way,
the defect percentage is calculated as a ratio and used to make an accurate defect
assessment. Lee et al . (2005) present a method for rust recognition on steel bridge
coating surfaces based on color images. They extract statistical variables of each
color channel as input for a multivariate discriminate function. The according
output is then used to determine the existence of rust defects. Validation results
show that this method can correctly classify coating images.