Australian Journal of Electrical and Electronics

Engineering

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tele20

Efficient crack detection and quantification in

concrete structures using IoT

Ajay Nair, Hemalatha R, P Sangeetha, Harish Kumar K, Dinesh Kumar P,

Inakota Sai Sahith & S. Radha

To cite this article: Ajay Nair, Hemalatha R, P Sangeetha, Harish Kumar K, Dinesh Kumar P,
Inakota Sai Sahith & S. Radha (2021) Efficient crack detection and quantification in concrete
structures using IoT, Australian Journal of Electrical and Electronics Engineering, 18:1, 43-57,
DOI: 10.1080/1448837X.2021.1914905

To link to this article: https://doi.org/10.1080/1448837X.2021.1914905

Published online: 26 Apr 2021.

Submit your article to this journal

Article views: 116

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tele20
AUSTRALIAN JOURNAL OF ELECTRICAL AND ELECTRONICS ENGINEERING
2021, VOL. 18, NO. 1, 43–57
https://doi.org/10.1080/1448837X.2021.1914905

Efficient crack detection and quantification in concrete structures using IoT

Ajay Naira, Hemalatha Ra, P Sangeethab, Harish Kumar Ka, Dinesh Kumar Pb, Inakota Sai Sahithb and S. Radhaa
a
Dept of Electronics and Communication, Sri Sivasubramaniya Nadar College of Engineering, Chennai, Tamilnadu, India; bDept of Civil, Sri
Sivasubramaniya Nadar College of Engineering, Chennai, Tamilnadu, India

ABSTRACT ARTICLE HISTORY

Now-a-days public transit in our country has a significant usage of bridges and tunnels. Received 1 September 2020
Maintaining the safety of such structures becomes the need of the hour. Cracking can invite Accepted 7 April 2021
sudden failures of concrete structures. Within recent years, there has been an increase in the KEYWORDS
use of image processing techniques as Non-Destructive Testing (NDT) method to detect Cracks; pattern classification;
defects and anomalies in such structures. Hence, this work presents an efficient image proces­ morphological processing;
sing model for identifying and quantifying the cracks in common structures using two algo­ NDT; IoT
rithms. The first algorithm uses pattern classification method to classify the identified
specimens as crack or not. This pattern classification algorithm is tested and evaluated with
the test specimens of structures having induced cracks. The second algorithm called morpho­
logical processing uses an efficient thresholding strategy to improve the detection accuracy of
the cracks. Here the aim is to detect even the faintest of cracks and magnify them. Thus, any
one of the algorithms can be adopted depending on the application. The developed algo­
rithms have been tested in real field at the final stage and achieved a detection accuracy of
nearly 96.37%and all the obtained images have been stored digitally on a web server.

1. Introduction approach but requires adaptation to the surface for

best results and involves a large number of compu­
Image Processing is a Non-Destructive Evaluation
tations. The image acquisition setup is also expen­
(NDE) method and has been developed to detect exist­
sive. It is desired to obtain the best results using
ing surface deteriorations including delamination and
a minimal cost setup. The methods provided in
voids in concrete. Traditionally, bridges have been
(Zhang et al. 2014a) give an insight into simpler
inspected by hammer sounding and/or chain drag,
crack detection and classification, with crack quan­
and visual inspections by qualified engineers and
tification; however, pre-processing requires high-
inspectors; however, these methods require a lot of
performance hardware equipment. The author in
field labour and lane closures, especially for bridge
(Hoang 2018) presented a novel thresholding
deck inspections (Yehia et al. 2007). Cracks are one
method to highlight the crack regions along with
of the major defects found in civil structures. They can
crack parameter estimation but failed to remove the
be viewed as a form of discontinuity that appears on
falsely detected artefacts. Simple edge detection
the surface of the structure and correspond to darker
ideas with false positives were also presented
regions of the image. Many of these cracks might be
(Dongna et al. 2012). Moreover, curve fitting mod­
harmless. However, the failure to detect harmful
els are used to identify the cracks. The Hough
cracks can become very costly over a period of time.
transform is a good tool for identifying line seg­
As more and more civil structures get built over the
ments in the data, and have been used in crack
years with older ones more prone to damage, it
detection systems (Nashat, Abdullah, and Abdullah
becomes unfeasible to manually inspect each structure
2011) but it does not work well when the cracks are
for defects. Hence, it is essential to develop an effective
not perfectly straight. The top hat transform is used
automatic crack detection model.
to detect cracks with a simple segmentation method
Crack detection can be performed by choosing an
in (Giakoumis, Nikolaidis, and Pitas 2006). Even
initial set of seed points and growing them (Oliveira
misidentified cracks are also corrected. However,
and LobatoCorreia 2017) but it takes a large number
the crack detection is not so efficient when they
of iterations to converge to the result. Moreover, it
cross the border between two regions. The detec­
does not calculate the crack properties. The road sur­
tion method used in (Landstrom and Thurley 2012)
face crack detection algorithm in (Gavilán et al. 2011)
was found to yield good results when cracks are
provides good results for different road surfaces and
long and connected in nature. It utilises the
performs connection operation by minimum distance

CONTACT Hemalatha R [email protected] Dept of Electronics and Communication, Sri Sivasubramaniya Nadar College of Engineering,
Chennai, Tamilnadu, India
© Engineers Australia
44 A. NAIR ET AL.

Figure 1. (a)Shrinkage cracks (b) Hairline crack.

method of filling the gaps between cracks but it can The paper is organised as follows. Information on the
turn up with relatively more number of false posi­ different types of crack is given in section 2. Section 3
tives. Methods for more specific applications such briefs about the proposed model. Section 4 provides the
as cracks in tunnels and underground pipes are performance evaluation and section 5 explains the IoT
also proposed which extract characteristic shapes (Internet of Things)-based monitoring. Finally, section 6
of such cracks for detection (Sinha and Fieguth concludes the paper with future work.
2006) (Seung-Nam, Jang, and Han 2007).
An effective segmentation method to identify crack
was proposed by Long et al. (Long, Shen, and Chen 2. Overview of cracks
2012). It was enabled through grey level correction Cracking can occur due to various reasons such as
and adaptive minimum error threshold but could not thermal stresses, chemical reaction, corrosion of rein­
classify cracks. Han et al. (Han et al. 2016) used an forcement or simply poor construction practices such
interactive genetic algorithm and determined the as poor concrete mix or rapid curing. Cracks can be of
edges of the cracks and calculated its geometric para­ different types such as alligator, transversal or long­
meters through intensive calculations. The combina­ itudinal (Ajagbe and Ojedele 2018) (Md. et al. 2018).
tion of Canny edge detector and support vector The other types are shrinkage and hairline cracks as
machine was used to detect cracks. This enhanced shown in Figure 1, which may result in leakage.
the ability to extract contours but struggled with Horizontal, vertical and diagonal cracks are charac­
cracks [Li et al. (Shi et al. 2016)]. Similarly, Canny terised by how they are oriented in the structure.
edge detection has been used along with fast Haar Alligator cracks occur as a V shape. Settlement cracks
transform algorithm for crack detection but, suffered may occur if there are problems with respect to the
with false crack detection. In addition, the method underlying ground.
could not extract crack pixels. The width of the cracks Manual introduction of cracks is performed in the
was also measured using labelling based on Red-Green laboratory to generate a data base. This involves mak­
-Blue (RGB) colour information (Hutchinson and ing of the test specimen in the desired shape as shown
Chen 2006). in Figure 2.
It is also to be noted that a single algorithm for Curing is done for about a month. After a couple of
crack detection is not always expected to give efficient hours of dryness it is kept in an oven for an hour for
results for all practical situations. Some structures hardening. Then, cracks are introduced by the univer­
might have a large number of cracks with lesser sal testing machine by applying the desired load as
major ones for repair purposes. Other structures may shown in Figure 3.
have few minute cracks but each crack might have to
be detected to ensure the safety of that structure.
Hence, the two proposed crack detection algorithms 3. Proposed system model
will cater correspondingly to the needs of the afore­ The proposed system can be visualised based on a flow
mentioned problems. The combination of algorithms diagram as shown in Figure 4.
proposed in this paper is more efficient in detecting The proposed system can be segmented into
even the finest cracks compared to the complex algo­ three major sections: the image processing, IoT-
rithms (Zhang et al. 2014a). Moreover, real-time mon­ based monitoring and the crack measurements
itoring methodologies are also illustrated to ensure and analysis section. The system has two algo­
seamless implementation. rithms for crack detection either of which can
 AUSTRALIAN JOURNAL OF ELECTRICAL AND ELECTRONICS ENGINEERING 45

Figure 2. (a) Sample specimen (b) Specimen undergoing load test.

Figure 3. Vertical crack introduced (a) on cube (b) on cylinder.

Figure 4. Proposed system model.

be used depending on the application, along with 3.1. Pattern classification algorithm
an IoT-based monitoring system. The first algo­
Pattern classification algorithm in Figure 5 is used in
rithm optimally detects the presence of true
this paper to classify a given specimen as crack or not
cracks and highlights them. This algorithm aims
a crack. It is applicable to a multi-class condition.
to eliminate maximum false positives generated
The distance of each specimen from each of the
due to inappropriate lighting conditions.
classes is measured and the class with minimum
The second algorithm highlights the presence of
distance is where the specimen belongs.
even the finest of cracks. This is useful in appli­
cations where even the slightest of cracking could The image database is created by collecting
be a potential threat for the future. This proposed a large number of similar images of various regions
system has been implemented in real time by of cracks. The acquired image is first converted to
hosting a web server that can monitor cracks greyscale. The quality of input image can be sig­
from a remote location. nificantly improved by preprocessing. It may
46 A. NAIR ET AL.

Figure 5. Classification algorithm flow diagram.

 AUSTRALIAN JOURNAL OF ELECTRICAL AND ELECTRONICS ENGINEERING 47

include the removal of noisy regions or improving pixel is replaced with the background pixel. This pro­
the lighting. Further denoising can be performed cess helps to remove foreground regions of the image.
using a median filter which essentially smoothens Dilation: The opposite process from erosion
the image. Median filtering preserves the image wherein if even one pixel of the structuring element
edges while removing the noise. A 3 × 3 kernel is overlaps with the foreground image, then the centre
used which takes the median of values in the pixel is replaced with the foreground pixel. This pro­
neighbourhood and replaces the centre pixel with cess helps to increase the number of foreground pixels.
it. A large kernel size is good for high noise den­ Closing: Closing is dilation followed by erosion.
sities but will excessively blur the image. Next, This morphological algorithm is represented in
binarization is used to convert pixel values to bin­ Figure 6.
ary as expressed in Equation (1). (1) Preprocessing: The image is filtered out by using
� Gaussian blur to remove unwanted noise components.
0ifx � T It is convolved with a Gaussian filter rather than a box
f b ðxÞ ¼ (1)
1ifx > T filter which replaces each pixel with a weighted aver­
age of the neighbouring pixels. It is used to remove the
where T is the selected threshold and f b ðxÞ is the
high-frequency components. Here, a 3 × 3 Gaussian
binarized image.
filter governed by Equation (2) is traversed over the
The selection of threshold is a critical step in order
image.
to optimally detect and separate cracks from the
� �
image. Here a statistical analysis is used to determine 1 ðx2 þ y2 Þ
the appropriate threshold. The greyscale peak value GðxÞ ¼ exp (2)
2πσ 2 2σ 2
(N) is obtained using the histogram and is taken as the
initial threshold. If the output is not satisfactory, the where G(x) is the expression for the Gaussian filter in
threshold is manually reduced by 1 pixel increments. two dimensions.
Generally, a threshold of around N-8 to N-10 was Larger the standard deviation (σ), greater is the
found to be suitable by experimentation. For a set of image blurring. It is important to preserve image qual­
images under a certain lighting, threshold will most ity along with reducing noise and so it is not advisable
likely not change. This is followed by image segmenta­ to go for larger values of σ. Here σ = 1 is found to suit
tion and labelling. Segmentation involves partitioning our requirements.
the image into its constituent parts. Labelling is done (2) Bottom hat transform: This is the subtraction of
on the image to identify possible regions of cracks. image closing by original image. This helps to obtain
This requires extraction of features pertaining to darkest regions of the image and remove most of the
cracks. The features used for crack detection are ratio background components (Zhang et al. 2014b). The
of major to minor axis and number of pixels. The ratio appropriate choice of the structuring element will
of major to minor axis for a typical crack is a large enhance the results. Generally, the shape of the struc­
value. The input image is given to the two-layer feed- turing element is chosen as close as possible to the
forward network to perform crack classification in shape that is to be highlighted. Hence, a rectangular
MATLAB (Matrix Laboratory Simulation Tool). It is structural element is used. If the size of this element is
trained using scaled conjugate gradient back propaga­ too small, it takes a longer time to completely traverse
tion. Data for classification is organised into the input the image thereby reducing its efficiency and can bring
matrix X and the target matrix T. Here, a selection of up more noise elements. When larger, the desired
15 neurons was able to provide good results. Half of regions are found to be thicker but can miss out on
the samples are used for training and remaining part some small-sized cracks. Here, a size of 15 × 15 is
for testing and validation. Its performance is moni­ found suitable to meet the desired requirements of
tored using cross entropy and confusion matrices. The highlighting the crack region as a clear output image.
final output is obtained after the phases of training, The operation can be depicted by the Equations (3) &
testing and prediction. (4) as follows.

A_ ¼ A � B ¼ ðA þ BÞ B (3)
3.2. Morphological processing algorithm
_
Anew ¼ A A (4)
The advantage of morphological image processing for
crack detection is that even the thinnest of cracks can _ Anew are the original, closed and output
Here A, A,
be detected and magnified so that it can be easily images, respectively, and B is the structuring element.
treated. + and– represent dilation and erosion operations
Erosion: The structuring element is moved over the respectively.{\displaystyle\ominus} =
entire image. If even one pixel of the structuring ele­ (3) Thresholding: The image thresholding is per­
ment overlaps with the background, then the centre formed using Otsu’s method (Yang et al. 2012). This
48 A. NAIR ET AL.

Image Acquisition

Gray Scaling

Gaussian Blur

Bottom Hat Transform

Otsu Thresholding

Noise Removal
Figure 6. Morphological algorithm flow diagram.

method involves selecting a particular threshold and connected and seem isolated, they can easily be
checking the number of pixels on either side. The aim removed as they behave as noise components.
is to find the threshold value where the sum of fore­
ground and background spreads is at its minimum. It
is suitable for images with a bimodal histogram dis­ 4. Performance evaluation
tribution. This helps to clearly visualise the crack Dice: Dice is the ratio of twice the intersection between
regions but along with it, a number of salt noises can the segmentation image and the ground truth to its
also occur. summation. Equation (5) shows the formula for the
(4) Noise Removal: The final image can be obtained Dice metric (Umeha, Hemalatha, and Radha 2018).
by removing all the noise components. Removal of
isolated or sparsely distributed pixels will effectively 2ðA \B Þ
Dice ¼ (5)
remove the noise, which requires the knowledge of AþB
pixel connectivity. The image is segmented into its where A is the segmented image and B is the ground
constituent components. When pixels are hardly truth.
 AUSTRALIAN JOURNAL OF ELECTRICAL AND ELECTRONICS ENGINEERING 49

Jaccard index: Jaccard index is the ratio between the 4.2. Morphological processing algorithm
intersection and union of the segmented result and the
The results obtained using morphological proces­
ground truth. Equation (6) shows the formula for the
sing algorithm are displayed as shown in Figure 8.
Jaccard index (Umeha, Hemalatha, and Radha 2018).
Improper thresholding due to poor contrast in the
image has resulted in poor output image quality in the
A \B
Jaccard ¼ (6) base algorithm. However, proposed morphological
A [B
processing gives a more magnified and lucid view of
where A is the segmented image and B is the ground the cracks. It is advantageous for detection of
truth. even minute cracks. The algorithm can be tweaked to
Emeasure: E-measure is defined by Equation (7) show even the smallest of defects. The drawback is that
and uses the enhanced alignment matrix φ to capture the detection of false positives cannot be avoided
the pixel-level matching and image-level statistics of easily. Figure 9 shows how the algorithm can be
a binary map. tweaked to show varying levels of the defects on
a surface.
1 X w X h
QFM ¼ ϕ (7)
w � h x¼1 y¼1 FMðx;yÞ 5. Iot-Based Monitoring
A complete crack detection and estimation model can
where h and w are the height and the width of the
be implemented using a Raspberry pi (RPi) setup. The
binary map, respectively, and ϕFM is the enhanced
monitoring system consists of an RPi and the Pi camera
alignment matrix (Fan et al. 2018).
(PiCam). The web server is hosted by the RPi and can
These provide a measure of closeness of output
be accessed by the user on any device by connecting to
image from the neural network with the ground
its IP address. A python program is used to implement
truth image, which can otherwise be termed as accu­
morphological image processing algorithm to process
racy of classification.
the images. The image of cracks can be uploaded from
the device or it can be taken remotely using the RPi and
uploaded to the server. The PiCam is also used to
4.1. Pattern classification algorithm
perform live streaming to get a real-time view of the
Crack detection has been performed using pattern inspected structure. The inputs as well as processed
classification method and various experimental images are displayed on the web server which is acces­
results are obtained. Figure 7 shows a portion of sible from any device connected to the internet.
images used in evaluating the system The Web server setup comprises of the following:
performance. (1) Web Server: Web Server is an integral part of the
From Figure.7 it is clear that superior results system being designed. A web server on RPi can be used
are obtained using the classification algorithm. to host a full website, or just used to display some. A Web
Large numbers of false positives are observed in server can be accessed through the RPi remotely through
the LDP (Local Directional Pattern) based imple­ a laptop by using SSH (Secure Shell). It is a cryptographic
mentation output and heavier noise corruption is network protocol that ensures secure data transfer
evident. Based on comparison with the ground between computer and RPi. Apache is a popular web
truth image, the Dice, Jaccard and E-measure server, which is a program that listens for server access
metrics and the accuracy of the classification are requests from Internet browsers and grants them if per­
calculated and provided in Table 1. mitted. The web servers can be accessed using its IP
From the pattern classification method of crack detec­
address.
tion, fairly good results are obtained with reasonable
accuracy. The primary advantage of this method is that (2) Internet Framework: Python is used to deploy
it is suitable for detection of even a large number of the web server on the RPi. The bottle web framework
cracks on the surface of the specimen with better accu­ architecture is used to make and enable web requests.
racy when compared to (Zhang et al. 2014a). Also it It is a simple and lightweight structure which comes
enables a very low percentage of false-positive detection. with inbuilt servers and can be used to perform var­
However, the disadvantage is that minute cracks might ious routing functions.
not be detected with good accuracy. The proposed algo­ (3) Live Streaming: The RPi camera can be used to
rithm is implemented in the on-board processor with scan the civil structure and the live video can be
lesser complexity. accessed using any device by connecting to the RPi.
50 A. NAIR ET AL.

Figure 7. A) Input Images b) LDP-based segmentation output c) Proposed classification algorithm-based segmentation output.
 AUSTRALIAN JOURNAL OF ELECTRICAL AND ELECTRONICS ENGINEERING 51

Table 1. Comparison of the performance evaluation metrics.

Evaluation Metrics
LDP Algorithm(Zhang et al. 2014a) Proposed Algorithm
Input Images Dice Jaccard Accuracy E-measure Dice Jaccard Accuracy E-measure
Image 1 0.6156 0.4446 80.48% 0.2830 0.9783 0.9576 95.12% 0.9976
Image 2 0.5265 0.3573 91% 0.3024 0.9598 0.9227 98.72% 0.9437
Image 3 0.5705 0.3991 84.84% 0.2803 0.9462 0.8980 93.94% 0.9273
Image 4 0.7872 0.6491 84% 0.6560 0.9268 0.8636 96% 0.9827
Image 5 0.5978 0.4263 80% 0.2942 0.9674 0.9368 96.67% 0.9862
Image 6 0.6459 0.4770 84.37% 0.8127 0.9481 0.9012 96.9% 0.9959
Image 7 0.6863 0.5224 91.17% 0.9303 0.8396 0.7235 94.1% 0.9825
Image 8 0.6971 0.5350 89.74% 0.8857 0.9034 0.8238 97.4% 0.9960
Image 9 0.0176 0.0089 78.94% 0.2542 0.8722 0.7733 94.7% 0.9772
Image 10 0.2338 0.1324 92.63% 0.4583 0.9274 0.8646 97.9% 0.9871

Figure 8. (a) Input Images (b) LDP-based segmentation output (c) Morphological processing algorithm-based segmentation
output.
52 A. NAIR ET AL.

Figure 9. (a) Original greyscale image (b) Output after morphological processing (c) Output after noise reduction.

Figure 10. Experimental Setup of RPi with (a) PiCam and (b) an LCD Display.

Figure 11. (a) Upload webpage (b) Database webpage.

From the live video, one can capture photos instantly user. This image is sent to the web server for proces­
and then use these photos for processing. sing. The captured image as well as the processed image
is displayed as output on the web server. The morpho­
logical processing algorithm has been implemented in
5.1. Real-time implementation Python in the RPi (https://www.raspberrypi.org). The
experimental setup is done as shown in Figure 10.
The experimental setup consists of RPi and PiCam.
This setup can be placed on any civil structure for Sample of the web pages is shown in Figure 11. The
continuous monitoring of the structure. The PiCam upload page asks the user to upload the image to be
can be used to take images whenever desired by the processed or the PiCam can be used. The database
 AUSTRALIAN JOURNAL OF ELECTRICAL AND ELECTRONICS ENGINEERING 53

Figure 12. Live streaming.

Figure 13. (a) Segmented Output 1 (b) Crack measurement results.

Figure 14. (a) Segmented Output2 (b) Crack measurement results.

page displays the results. These operations can be done 5.2. Measurements and analysis
remotely from any location.
Once the cracks have been identified and processed,
To obtain a real-time view of the structure to be
the next step is to analyse it by performing certain
inspected, live-streaming feature has been provided for
calculations to determine the potential harm that the
continuous monitoring. This process is shown in
crack could possibly impart to the structure under test.
Figure 12.
54 A. NAIR ET AL.

Figure 15. (a) Segmented output 1 (b) Confusion matrix.

Figure 16. (a) Segmented output2 (b) Confusion matrix.

Figure 17. (a) Segmented output 3 (b) Confusion matrix.

Also it provides the accuracy of crack detection. Thus, repaired. Such quantitative parameters obtained make
once all the parameters are obtained one has to take it easier to ensure a cost-effective treatment of the
a decision on when the structure becomes ready to be affected structure.
 AUSTRALIAN JOURNAL OF ELECTRICAL AND ELECTRONICS ENGINEERING 55

Figure 18. (a) Segmented output 4 (b) Confusion matrix.

Figure 19. (a) Segmented output 5 (b) Confusion matrix.

Table 2. Metric values.

Metric Image 1 Image 2 Image 3 Image 4 Image 5 Image 6 Image 7 Image 8 Image 9 Image 10
Accuracy 95.12% 98.72% 93.94% 98.3% 96.6% 96.9% 94.1% 97.4% 94.7% 97.9%
True Positive 22 20 23 35 19 11 12 13 12 36
False Positive 2 1 2 0 1 0 2 1 1 2
False Negative 0 0 0 1 1 1 0 0 0 0
True Negative 17 57 7 24 39 20 20 25 6 57
Precision 0.917 0.952 0.92 1 0.95 1 0.86 0.93 0.92 0.95
Recall 1 1 1 0.972 1 0.92 1 1 1 1

5.2.1. Crack dimensions the various connected components. Distance between

Measurement of crack parameters is essential to get the the two farthest points in a connected component gives
crack characteristics and adopt a certain treatment the crack length. Crack Area is determined by the num­
mechanism. The number of pixels is obtained from ber of pixels in the crack. A large crack area but a small
56 A. NAIR ET AL.

length means that the width of the crack is large. Crack Funding
Average width is given by the perpendicular distance
This work was supported by the SSN Trust under Student
between the two end pixels of the midpoint of the line Internal Funding Project;SSN Trust;
joining the two farthest pixels. A larger crack width
usually implies that the associated structure is more
severely affected. Each of these measurements is done Notes on contributors
for the largest crack, which has the potential to cause Ajay Nair obtained his Bachelor’s degree in Department of
maximum harm. These measured parameters give an Electronics and Communication Engineering from Sri
idea of how large the crack is and if it is of immediate Sivasubramaniya Nadar College of Engineering,
concern. Figure 13 and Figure 14 show the measure­ Tamilnadu, India.
ment results of the largest crack for the particular image. R. Hemalatha is an Associate Professor in the Department
of Electronics and Communication Engineering, Sri
5.2.2. Evaluation metrics Sivasubramaniya Nadar College of Engineering,
Tamilnadu, India. Her area of interest includes Wireless
The metrics which contribute to the overall analysis
Multimedia Sensor Networks, Image Processing, and IoT.
of the crack detection method are calculated in this
section for the five images which are used in the P. Sangeetha is an Associate Professor in the Department of
Civil Engineering, Sri Sivasubramaniya Nadar College of
classification algorithm. The processed images for Engineering, Tamilnadu, India. She completed her B.E
few samples are shown in Figures 15-19 along with (Civil), MS (Research), and Ph.D from college of
their confusion matrices. This will enable the cal­ Engineering, Guindy, Anna University. Her area of interest
culation of the metrics. Table 2 summarises the includes composite space truss, steel concrete composite,
metric values obtained from the test images. cold formed steel sections and finite element analysis.
From Table 2 it is inferred that the proposed algo­ Harish Kumar Kobtained his Bachelor’s degree
rithm achieves a detection accuracy of nearly 96.37% in Department of Electronics and Communication
on an average with better precision. The accuracy is Engineering from Sri Sivasubramaniya Nadar College of
Engineering, Tamilnadu, India.
calculated with reference to the manually built ground
truth. P. Dinesh Kumar , obtained his Bachelor’s degree in
Department of Civil Engineering from Sri
Sivasubramaniya Nadar College of Engineering,
6. Conclusions and future work Tamilnadu, India.
Sai Sahith I, obtained his Bachelor’s degree in Department
The presented paper elaborated on two techniques of of Civil Engineering from Sri Sivasubramaniya Nadar
crack detection and backed it up with a real-time crack College of Engineering, Tamilnadu, India.
detection model. The first technique did crack detec­
S. Radha is the Professor and Head of the Department of
tion by measuring how well the cracks correlate with Electronics and Communication Engineering, Sri
standard parameters. This method ensures high crack Sivasubramaniya Nadar College of Engineering,
detection accuracy and will significantly reduce false Tamilnadu, India. Her area of interest includes Wireless
positives. The second technique used morphological Sensor networks, energy harvesting devices, MEMS,
processing which detects and highlights even the faint­ NEMS, IoT and cognitive radios.
est of cracks. This method is particularly useful when
even the smallest of cracks is of due concern. References
Depending on the application, either of the processing
techniques can be used. Further the hosting of a web Ajagbe, W. O., and O. S. Ojedele. 2018. “Structural
Investigation into the Causes of Cracks in Building
server enables remote monitoring of a structure using and Solutions: A Case Study.” American Journal of
IoT. Also the quality of images that are taken can be Engineering Research (AJER) 7 (8): 152–160.
improved with the help of a camera with higher reso­ Dongna, H., T. Tian, H. Yang, S. Xu, and X. Wang,“Wall
lution. This monitoring system can be integrated with Crack Detection Based on Image Processing” 2012 Third
a mobile robot to allow scanning of the entire struc­ International Conference on Intelligent Control and
Information Processing, July 15–17, Dalian, China,
ture. Moreover, the correlation between the size and
2012, pp. 597-600, 10.1109/ICICIP.2012.6391474.
type of cracks can be examined to comprehend the Fan, D.-P., C. Gong, Y. Cao, B. Ren, -M.-M. Cheng, and
complexity and severity associated with it in real A. Borji, “Enhanced-alignment Measure for Binary
scenarios. Foreground Map Evaluation”, in Proc. Twenty-Seventh
International Joint Conference on Artificial Intelligence
(IJCAI-18),July 2018, Stockholm, Sweden, pp.698–704.
Disclosure statement Gavilán, M., D. Balcones, D. F. Oscar Marcos, M. A. Llorca,
I. P. Sotelo, M. Ocaña, P. Aliseda, P. Yarza, and
No potential conflict of interest was reported by the A. Amírola. 2011. “Adaptive Road Crack Detection
author(s).
 AUSTRALIAN JOURNAL OF ELECTRICAL AND ELECTRONICS ENGINEERING 57

System by Pavement Classification.” Sensors 11: Refinement”, 25th European Signal Processing
9628–9657. doi:10.3390/s111009628. Conference (EUSIPCO), Kos island, Greece, 2017,
Giakoumis, I., N. Nikolaidis, and I. Pitas. 2006. “Digital pp.2080-2084
Image Processing Techniques for the Detection and Seung-Nam, Y., J.-H. Jang, and C.-S. Han. 2007. “Auto
Removal of Cracks in Digitized Paintings.” IEEE Inspection System Using a Mobile Robot for Detecting
Transactions on Image Processing 15: 178–188. Concrete Cracks in a Tunnel.” Automation in
doi:10.1109/TIP.2005.860311. Construction 16: 255–261. doi:10.1016/j.
Han, F., S. Zhao, L. Zhang, and J. Wu. 2016. “Survey of autcon.2006.05.003.
Strategies for Switching off Base Stations in Shi, Y., L. Cui, Z. Qi, F. Meng, and Z. Chen. 2016.
Heterogeneous Networks for Greener 5G Systems.” “Automatic Road Crack detection Using Random
IEEEAccess 4 (August): 49594973. Structured Forests.” IEEE Trans. Intell. Transp.Syst 17
Hoang, N.-D. 2018. “Detection of Surface Crack in Building (12, December): 34343445. doi:10.1109/
Structures Using Image Processing Technique with an TITS.2016.2552248.
Improved Otsu Method for Image Thresholding.”
Sinha, S. K., and P. W. Fieguth. 2006. “Automated Detection
Hindawi Advances in Civil Engineering 2018, pp.1-10.
of Cracks in Buried Concrete Pipe Images.” Automation in
Accessed 15-April-2021
Construction 15 (January): 58–72. doi:10.1016/j.
https://www.raspberrypi.org
autcon.2005.02.006.
Hutchinson, T. C., and Z. Chen. 2006. “Improved Image
Analysis for Evaluating Concrete Damage.” Journal of Umeha, M., R. Hemalatha, and S. Radha, “Structural Crack
Computing in Civil Engineering 20 (3, May): 210–216. Detection Using High Boost Filtering Based Enhanced
doi:10.1061/(ASCE)0887-3801(2006)20:3(210). Average Thresholding”, in the Proceedings of 2018
International Conference on Communication and Signal
Landstrom, A., and M. J. Thurley. 2012. “Morphology-based
Processing (ICCSP),Chennai, India, 2018, pp. 1026-1030,
Crack Detection for Steel Slabs.” IEEE Journal of Selected
10.1109/ICCSP.2018.8524284.
Topics in Signal Processing 6: 866–875. doi:10.1109/
JSTSP.2012.2212416. Yang, X., X. Shen, J. Long, and H. Chen. 2012. “An
Improved Median-based Otsu Image Thresholding
Long, J. W., X. J. Shen, and H. P. Chen. 2012. “Interactive
Algorithm.” AASRI Procedia 3: 468–473. doi:10.1016/
Document Image thresholding Segmentation Algorithm
j.aasri.2012.11.074.
Based on Image Regions.” Journal of Computer Research
Yehia, S., O. Abudayyeh, S. Nabulsi, and I. Abdelqader.
and Development 49(7):14201431. July.
2007. “Detection of Common Defects in Concrete
Md., S. A., B. M. AmrulKaish, C.-O. Woon, and Bridge Decks Using Non-destructive Evaluation
S. N. Raman. 2018. “Early-Age Cracking in Concrete: Techniques.” Journal of Bridge Engineering 12: 215–225.
Causes, Consequences, Remedial Measures, and doi:10.1061/(ASCE)1084-0702(2007)12:2(215).
Recommendations.” Applied Sciences 8: 1730. Zhang, W., Z. Zhang, D. Qi, and Y. Liu. 2014a. “Dapeng Qi
doi:10.3390/app8101730. and Yun Liu, “Automatic Crack Detection and
Nashat, S., A. Abdullah, and M. Abdullah, “A Robust Crack Classification Method for Subway Tunnel Safety
Detection Method for Non-uniform Distributions of Monitoring”.” Sensors 14 (10): 19307–19328.
Coloured and Textured Image,” inProc. IEEE Int. Conf. doi:10.3390/s141019307.
Imag. Syst. Tech. (IST), Batu Ferringhi, Malaysia, May Zhang, W., Z. Zhang, D. Qi, and Y. Liu. 2014b.
2011, pp. 98–103. “Automatic Crack Detection and Classification
Oliveira, H., and P. LobatoCorreia, “Road Surface Crack Method for Subway Tunnel Safety Monitoring.”
Detection: Improved Segmentation with Pixel-based Sensors 14: 19307–19328.