Computer Vision and Pattern Recognition Technologies for Construction - Mobile and Pervasive Computing in Construction - page 197

isolated crack pixel is marked white, and non-crack pixels are marked black. These
methods use special visual characteristics of cracks in images and adopt various
image processing techniques, such as wavelet transforms, threshold, and edge
detection (Canny edge detector, Sobel edge detector, Fourier transform, fast Haar
transform), to extract crack pixels from the image background. Cheng et al . (2003)
detect cracks by simply thresholding the concrete surface image. The threshold
value is determined based on the image's mean and standard deviation values.
Abdel-Qader et al . (2003) presented a comparison of edge detection techniques
with respect to concrete bridges images and found that the fast Haar transform is
most reliable. However, in these global processing methods detection accuracy is
affected by image noise. To address the problem of image noise, Yamaguchi and
Hashimoto (2009) propose a fast and scalable local percolation-based image
processing method that considers crack connectivity among neighboring image
pixels. Also, Sinha and Fieguth (2006) introduce two crack detectors that consider
relative statistical properties of adjacent image regions. These two detectors
are applied in four directions (0 ,45 ,90 , 135 ) to identify crack pieces in
buried concrete pipes; then a linking and cleaning algorithm is used to connect
crack pieces.
The third category contains methods that use crack maps to retrieve crack
properties like length, maximum width, average width and orientation. Yu
et al . (2006) calculate the length, width and orientation of cracks through a
graph search; however, their method required the start and end points of the
crack to be manually provided first. Chae et al . (2003) use an artificial neural
network to retrieve crack properties. Zhu et al . (2010) propose a method that
creates topological skeletons of cracks through binary image thinning and calcu-
lates the distance field of crack pixels in the map using a distance transform.
According to skeleton configurations and the distance values of crack pixels, crack
properties (width, length, orientation and location) are retrieved with an average
error of 3%.
10.3.2 Air pockets
As a result of the entrapment of air bubbles during the concrete placement and
consolidation process, air pockets reduce the concrete's strength, increase its
permeability, decrease its bond to the reinforcement and severely undermine
the desired appearance and visual uniformity of architectural concrete. According
to the distinctive near-circular shape of air pockets, Zhu and Brilakis (2008) have
created a spot filter which is the combination of three concentric, symmetric
Gaussian filters. Subsequently, a concrete surface image is convolved with this filter
and high response values, the maximum response values in local areas, are expected
in the places where air pockets exist. In this way, air pockets whose size is similar to
that of the filter can be detected directly by locating the high response values. A
multiscale representation of the input image (image pyramid) is used to detect air
pockets of different sizes. The properties of air pockets (number, size, and
occupation area) are subsequently calculated. This method can correctly detect
87% of air pockets that vary in size.