See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/384429783

AI and Applications in Civil Engineering _Module 5_ Computer Vision

Chapter · September 2024

CITATIONS READS

0 49

1 author:

Jayaram M.A
RASTA - Center for Road Technology VOLVO Construction Equipment Campus
196 PUBLICATIONS 1,506 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jayaram M.A on 28 September 2024.

The user has requested enhancement of the downloaded file.

AI and Applications in Civil Engineering
Module 5

COMPUTER VISION

RASTA
Center for Road Technology
VOLVO-CE CAPMUS
Bengaluru

Dr.M.A.Jayaram
Professor
 CONTENTS
Introduction to Computer Vision
Phases of CV
Image Acquisition Methods
Tools for Image Acquision
CV methods
Intuitionistic Understanding of Related algorithms
Applications of CV in Civil Engineering
Module End Questions
References
AI and Applications in Civil Engineering

Module 5
Computer Vision and Applications
1. Introduction
Computer vision (CV) is a branch of AI and also an interdisciplinary field founded on science.
It is supported by several other technologies like digital signal processing, artificial
intelligence, machine learning, and higher order mathematics. It is all about making computers
to get the grasp or understand the visual data in terms of images or videos as inputs. The level
of understanding vaguely mimics that of human visual system. As of today, the range of
applications of computer vison could be far from anybody’s imagination. In a nutshell, the
predominant applications include traffic monitoring and management, surgical automation,
automation in industry, automated driving, eye and head tracking in e-commerce, inspections
and monitoring during manufacturing stages, sports analysis, gesture recognition, general
scientific vision systems, object recognition, people tracking, safety monitoring, security and
biometrics, three-dimensional modelling, web and cloud applications, and many more.

In recent years, computer vision is associated closely with architectural engineering and civil
engineering. This trend is fast growing particularly in civil engineering related activities. A
plethora of applications relate to construction and management of the infrastructural facility. It
is well known that the construction industry is marred by poor construction quality,
mismanagement at construction sites, exorbitant overhead costs, injuries and health issues of
construction workers, improper measurements of structural distress, mismatching in schedules,
and other clutters, disorders that are very intrinsic. Justifiably to address this, the researchers
have explored and have been exploring the potentials of computer vision to effectively deal
with performance monitoring, resources management and safety assurances.
In the last one-and-a-half-decade copious literature is accrued. A sizable volume of literature
hinges on automated health monitoring of structures. The heart of the matter is CV is applied
for the tasks that are directed towards construction, monitoring of structural health, movement
of materials, anomaly detection and many more. A graphical abstract of CV and their various
applications are provided in figure 5.1.

2. Steps of CV
CV simply told is precisely a seven-step procedure segregated under three phases. A schematic
diagram is depicted in Fig. 5.2. The first phase involves preliminary processing of the image
or video. Feature extraction focuses on reducing the vast amount of digital information from
the image to relevant features. This is accomplished through dimensionality reduction
techniques such as principal component analysis (PCA), independent component analysis
(ICA), linear discriminant analysis (LDA), locally linear embedding (LLE), and t-distributed
stochastic neighbor embedding (t-SNE). Depth perception, or stereopsis, involves using
multiple cameras to capture the depth of objects within the scene. By matching objects in two
images and analyzing the differences in their relative positions, the depth of the objects can be
estimated. When more than two cameras are involved, the process is referred to as multi-view
stereopsis.

1 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

In the next phase, advanced image processing techniques are applied. During segmentation, the
image is divided into distinct regions based on pixel intensity, allowing for the identification
of damaged structural components, the measurement of cracks, and the detection of harmful
materials. Image tracking involves detecting specific regions of interest and monitoring these
regions over time for potential changes. This technique can identify symmetry, hidden patterns,
and various differences between the region of interest and other parts of the image. Visual
tracking further enables the monitoring of object movement (such as people, materials, and
construction equipment) and provides estimates of future positions of the tracked targets.
The third and final phase involves higher-level processing of images or videos. Image
registration identifies spatial transformations occurring within a set of images, ensuring they
align within a common reference frame. This step is crucial when combining data from multiple
images and is particularly time-intensive when working with 3D images. The final step is object
recognition, where the system identifies objects, people, or entities and analyzes intricate
details. After recognition, the objects are classified or labeled based on pixel intensities and
their distribution patterns.

Fig. 5.1 : Graphical abstract of CV and its applications in civil Engineering [7]

3. Acquisition of Image
The visual data may be availed or acquired in three forms, 2D images which includes video,
3D images, and point clouds to capture objects to represent objects or space. In the fields of
image processing and computer vision, image acquisition refers to the process of obtaining an
image from a source, typically hardware devices such as cameras or sensors. This step is
crucial, as it initiates the entire workflow; without capturing an image, the system cannot
perform any further processing. The image obtained at this stage is typically raw and has not
undergone any processing. As the quality of data is very crucial to object registration and
recognition, the parameters and their thresholds are clearly defined in most of the CV tasks and

2 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

particularly with reference to cloud points. The specifications for the capturing devises are
readily available in references with related real time applications. Given the current day
sophistication, a smart phone of high resolution would suffice for preliminary data collection.
A broad overview of data acquisition methods is given in the following paragraphs

Fig. 5.2 . CV as a three-phase process [ 1]

The specifications for the capturing devises are readily available in references with related real
time applications. Given the current day sophistication, a smart phone of high resolution would
suffice for preliminary data collection. A broad overview of data acquisition methods used by
researchers and the benefits and lacunas are presented in Table 5.1. The details of the processes
involved are elaborated in the references mentioned
Table 5.1 .Visual data formats and related equipment.
Visual Data Format Capturing Devices
Image 2D /Video Digital camera of high resolution, camcorder, digital video camera
smart phone, surveyance camera. Ex. Apple i pod, iphone, LGE Nexus
5 etc. . ..

Image 3D Stereo camera, RGB-D Camera (for depth sensing). Ex.Bumblebee

XB3- IEEE-1349b, MS Kinnet/ Assure Kinnet etc. . .

Cloud Points Laser Scanner. Ex. Triamble GX 3D spatial camera, Leica Geosystems
3D laser etc..

3.1 Mobile Optical and Static Sensors

Portable mobile sensors are either held in hand or mounted on hardhats. Instances of they are
being held my mobile robots is reported. They are found to be vulnerable for adverse

3 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

environmental conditions. Some have recommended for the augmenting intelligent algorithms
to deal with unforeseen or uncontrollable parameters. They are shown in figure 5.3 and figure
5.4.

Fig. 5.3 Mobile optic sensor Fig. 5.4 Static optic sensor

3.2 Stationary Camera

Is the camera, fixed at a location with fixtures. Can collect visual data which is devoid of blur
or haze. However, they are susceptible to mild jerks and shakes owing to wind. The cameras
must also be positioned at an appropriate distance from the scene to compensate for the sensors'
range limitations. These machine vision cameras utilize the Gigabit Ethernet (GigE) Vision
interface standard to transfer image data. It defines a standard method for transmitting high-
quality images and control data over Gigabit Ethernet networks. Figure 5.5 shows such a
camera.
3.3 Surveillance Systems
Generally mounted at an elevated and farther off places for obtaining a perfect field of view.
Very ideal for visualizing the progress of work, monitoring quality of work, the safety and
security issues. Closed circuit television is widely used. Another advantage is noninterference
of construction managers, as they can take a far field view. Particularly ideal for visualizing
and monitoring construction activities. Shown in Figure 5.6

Fig. 5.5 Stationary Camera Fig 5.6. Surveillance Camera

3.4 Unmanned Arieal Vehicles (UAV)
A state-of-the art technology that is fast gaining acceptance due to its flexibility in bisecting or
targeting perspectives and elevations. Due to their ability to cover wide areas and capturing
ultra-high-resolution aerial images at low cost, drones became an extremely important source

4 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

of visual information. The images obtained through this are of moderate quality owing to
possible vibrations and dislocations. Thus, the images demand huge preprocessing. Shown in
figure 5.7.

Fig 5.7 Unmanned arial vehicle Fig. 5.8 Sterio vision

3.5 Sterio Vison and Laser Scanners

This system will capture dynamic scenes. They use multiple monocular cameras. Laser
scanners can capture 3D points with adequate accuracy not cost effective though. Laser
Scanners are non-contact devices that capture millions of discrete data points to measure an
object or space using laser infrared technology that produces detailed 3D images in minutes.
The images comprise millions of 3D data points, known as a point cloud. The processes like
scanning and certain calibrations require trained personnel. As huge volume of the data
involved, the interpretation will be tough. Sterio vison device is shown in figure 5.8. A laser
scanner is shown in figure 5.9.

Object /scene to be imaged

RGB Infrared Camera 3D scanner

Combining data from multiple sensors to create

a comprehensive image

Hybrid image/ processed data

Fig..5.9. 3D- laser scanner. Fig. 5.10. Hybrid acquisition system

3.6 Hybrid Systems

Hybrid imaging systems entail five functions: capture, input, processing, storage, and output.
Image capture through conventional photography is the analog component of a hybrid imaging
system, while the remaining functions constitute the digital component. hybrid system must
provide four general capabilities: high quality image input and output, optimum color
conversion, powerful image processing, and efficient image data handling. In hybrid
acquisition systems, Hardware configurations can be customized to adopt multiple gadgets such
as cell phone, camcorder, light detection ranging system can be merged with unmanned arial
vehicle to optimize the range of shooting and efficiency. A hybrid image acquisition system
combines multiple imaging modalities to capture comprehensive information from a scene or
object. These systems typically integrate various types of sensors, such as RGB cameras,

5 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

infrared cameras, LiDAR, or other spectral sensors, to produce a more detailed and accurate
representation of the subject. A typical block diagram of such a system is shown in figure 5.10.
Hybrid image acquisition systems are widely used in fields such as remote sensing, medical
imaging, autonomous driving, and construction monitoring, where integrating data from
multiple sensors provides more comprehensive and actionable information.

4. Processing of Acquired Images

Preprocessing of images that are obtained through any of the acquisition techniques discussed
above are to be subjected to preprocessing. The kind of method of acquisition can substantially
avoid noise, blurring and other undesirable aberrations. However, it is a matter of degree.
Filtering the image by linear or nonlinear filters, clipping, and framing are common
preprocessing techniques. Images are also transformed from one color image model to other,
generally from RBG ( RED , BLUE, GREEN ) to HSY (Hew Saturation Value) model. The
HSV model offers a numerical representation of an image corresponding to the color names it
contains. In this model, hue is measured in degrees ranging from 0 to 360. Transforming an
image from the RGB model to a grayscale model becomes necessary when the RGB format is
insufficient for extracting color information, or when distinguishing primary colors using
specific filters is challenging. Images captured under suboptimal lighting or poor illumination
can be improved using a technique called histogram equalization. Key steps involved in
processing an image from its raw state to a more refined form are outlined below

4.1 Image/ Scene Reconstruction

Scene reconstruction refers to the process of recreating a 3D model of a scene from 2D images.
Think of it like turning a photo back into a 3D environment. By using techniques such as
structure-from-motion or stereo vision, multiple images taken from different angles can help
determine the position of objects and the depth of the scene. This step is crucial when the spatial
relationships between objects in a scene need to be analysed.
The necessity for scene reconstruction arises from the fact that the depth images that are
generated by consumer digital cameras are laden with noise and missing details particularly
when surfaces are bright, transparent, shiny, or distant away from the camera [41]. Scene
reconstruction will result in 3D models that are complete and highly detailed in geometry and
provide high-quality surface textures. This is a two-step process involving local and global
registration of features of interest. Iterative closest point (ICP) and principal component
analysis algorithms are widely used to obtain image registration with required fineness or
granularity. In figure 5.11, a scene reconstruction is evident, the bits and pieces as appeared in
the original image has been reconstructed to form two separate built ups.

6 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

Fig. 5. 11. Original image at centre, reconstructed at the sides.

4.2 Object Parsing

Object parsing is breaking down an image into its components, like identifying different
regions and labelling them (for example, separating the sky, buildings, and cars in a city image).
It's like teaching a computer to look at a picture and recognize different objects in it, even if
they overlap or are partly hidden. This helps in segmenting the image so that each part can be
treated separately for further analysis. Object parsing will result in labelling of building,
ground, and any other construction entity to each pixel of the image.
Also called semantic segmentation is carried out through parametric and nonparametric
approaches. As far as implementation of this step in civil engineering applications, the posture
estimation of construction workers, rigid structure’s location, the movement sand location of
earth moving equipment, and any such entities. The output of this step is a skeletal model of
the entities of interest. Figure 5.12 brings how parsing of different objects in a construction site
scene is parsed. This will help the process of object recognition and libelling.

Fig. 5.12. Scene parsing to recognize the construction site objects

7 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

4.3 Object Recognition

In this step, one or more targets or objects under study are detected and identified. This is a
crucial step as it enables the machine vision understanding. Object recognition may extend
itself to movements, activities of the object in dynamic vision analysis. State-of-the-art
methods for object detection from generic classes are mainly based on deep convolution
networks. The framework of methods partitions the detection into several steps, pre-training,
fine-tuning, training with a classifier, and regression based on bounding box method. This well-
organized sequence of steps has resulted in increased performance in terms of accuracy and
thus becoming popular method to be implemented by other researchers. Algorithms have also
been proposed to detect category and location information.
Pertinent to say that improved models, such as fast region-based convolution networks
(RCNN). Further, the objective set forth was to develop method that could speed up the
processes of object detection [50]. As far as civil engineering applications are concerned, object
activities include, earth moving operations, mounting and dislodging formwork and tying of
rebars are common to various construction projects can stand to be examples of object
activities. Figure 5.13 highlights the identification of objects in a construction site.

Fig. 5.14. Detection of objects in a construction site scene.

5. CV Methods and Algorithms

Computer vision-based methods and algorithms allow machines to interpret, process, and
analyze visual data from the world. These methods rely on various algorithms that enable a
computer to "see" images or videos, extract useful information, and make decisions. Some of
the predominant methods are explained below.
5.1 Image Segmentation
Image segmentation divides an image into different regions or objects to make it easier for
computers to process. Each segment is analysed individually. Algorithms like thresholding,
edge detection, and region-growing are used for this purpose. As an example, in a satellite
imagery or aerial photographs algorithms can differentiate land-use areas, identify roads, or

8 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

track the extent of construction zones. Figure 5.15 shows the segmentation of various objects
in an image carrying construction scene.

Fig.5.15. Segmentation of objects in construction site for visual understanding

The intuitive elaboration the segmentation algorithms is made in Table 5.2.
Table 5.2 Intuitive explanation on segmentation algorithms
Algorithm What it does? How it does?
Thresholding: Converts grayscale images to binary by separating foreground and background.
Compares pixel intensity to a threshold value; pixels above the threshold are
classified as one region, those below as another. The pixels above a certain intensity
value (the threshold) are set to white (1), while those below are set to black (0).
Thresholding is commonly employed to detect objects in controlled environments,
such as identifying road markings or isolating pavement cracks from the
background in civil engineering images. This technique simplifies image analysis
by converting grayscale images into binary images, making it easier to distinguish
between objects of interest and their surroundings.
Edge Edge detection is a technique used to identify the boundaries or edges of objects
Detection: within an image. It works by detecting areas where there are sharp changes, or
discontinuities, in pixel intensity, which often correspond to the edges of objects.
By focusing on significant shifts in pixel values, edge detection highlights the
boundaries that define the shapes and structures in the image Two algorithms are
prevalent.
I. Gradient-based methods: These calculate the gradient (rate of change) of pixel
intensities across the image. If the gradient is large, the pixel lies on an edge
ii. Canny Edge Detection: One of the most widely used methods, Canny edge
detection first smooths the image to reduce noise, then calculates gradients, applies
non-maximum suppression to thin out edges, and uses double-thresholding to
identify strong and weak edges.
Region Region growing is a segmentation technique that groups neighbouring pixels based
Growing: on predefined criteria, such as similar pixel intensity or color. It starts with a seed
pixel and expands the region by including neighbouring pixels that share similar
properties.
• The algorithm begins by selecting one or more seed points, typically in regions
that meet a certain criterion (e.g., intensity within a certain range). These seeds
act as starting points for growing regions.

9 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

• The algorithm then examines the neighbouring pixels of the seed point. If these
neighbouring pixels have similar characteristics (such as similar intensity or
colour), they are added to the region.
• This process continues iteratively, checking the neighbours of each newly
added pixel, until no more pixels can be added (i.e., the region stops growing).

5.2 Feature Extraction

Feature extraction involves identifying important characteristics of an image, such as edges,
textures, and corners, which are essential for tasks like recognition or classification. Algorithms
such as SIFT (Scale-Invariant Feature Transform) and Harris Corner Detector help extract
robust features from images. A brief explanation of these algorithms is made in table 5.3.
As an example, in bridge inspection, feature extraction techniques are used to detect cracks or
deformations in the structure by analysing surface textures and irregularities.
Table 5.3. Intuitive explanation of feature extraction algorithms
Algorithm What it Does? How it Does?
SIFT: SIFT is an algorithm used to detect and describe local features in
(Scale Invariant Feature images. It’s particularly useful because it is invariant to scale,
Transform) rotation, and partially invariant to changes in illumination and
viewpoint. SIFT extracts distinctive key points (features) from an
image that can be used for matching objects between different
images or for object recognition.

SIFT features are highly distinctive, so once keypoints from two

images are detected, the algorithm matches them based on their
descriptors, allowing for robust object recognition or image
stitching.
Major tasks: Object recognition, image stitching, panorama creation,
and feature matching

Harries Corner Detection: The Harris Corner Detector is used to find corners (or interest points)
in an image. Corners are points where the intensity changes
significantly in multiple directions, which makes them ideal features
for tracking, recognition, or matching tasks.

Major tasks: Image matching, motion tracking, and corner detection

in structural analysis

5.3 Object Detection and Recognition

Object detection involves identifying and locating objects in an image. Object recognition goes
one step further by not just locating but also classifying the detected objects. Algorithms like
YOLO (You Only Look Once) and Faster R-CNN (Region-based Convolutional Neural
Networks) are commonly used. As an example, these algorithms can detect and recognize
construction equipment on-site, detecting cracks or structural issues in buildings, or identify
traffic signs in road networks given the respective photographs or videos. These algorithms are
briefly explained in Table 5.4.

10 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

5.4 Optical Flow

Optical flow tracks the movement of objects, pixels, or features across a sequence of images
(videos). This helps estimate motion and dynamic changes. Methods like the Lucas-Kanade or
Horn-Schunck algorithms are used for optical flow.
For an example, optical flow could be implemented for monitoring traffic flow on highways,
for analysing pedestrian movement in public spaces, for measuring the dynamic behaviour of
structures under stress (such as bridges or skyscrapers in the wind). A brief explanation about
the related algorithms is done in Table 5.5.
Table 5.4. Intuitive explanation of object detection and recognition
Algorithm What it does? How it does?
YOLO YOLO is an object detection algorithm that identifies and localizes multiple
(You Only Look objects in an image in a single pass. Unlike traditional methods that use a
Once) two-stage approach (region proposal followed by object detection), YOLO
treats object detection as a regression problem, predicting both bounding
boxes and class probabilities directly from the image in real time.
• The image is divided into a grid (e.g., 7x7 grid).
• For each grid cell, YOLO predicts a fixed number of bounding boxes
(typically 2) and the confidence score for each box. The confidence score
represents the probability that an object exists in that box.
• YOLO also predicts the class probabilities for objects in each grid cell.
These probabilities are multiplied by the confidence score to get the final
prediction of the object.
• Unlike traditional object detectors, YOLO performs detection in a single
forward pass through the neural network. This makes it incredibly fast
and efficient.
• Each grid cell predicts bounding boxes and corresponding class labels,
and these predictions are refined to output accurate bounding box
coordinates.
• YOLO filters out overlapping boxes, keeping only the ones with the
highest confidence scores.
A typical output of YOLO algorithm is presented in figure 5.15

R-CNN: Faster R-CNN is a two-stage object detection framework. It is an extension

(Region based of the R-CNN family, which includes region proposals and convolutional
Convolutional neural networks (CNNs).
Network) • The input image is passed through a deep convolutional neural network
(usually a pre-trained backbone like ResNet or VGG). The CNN extracts
feature maps from the image.
• Faster R-CNN generates potential regions of interest (ROIs) where
objects might be located.
• It proposes candidate object regions (anchors) at various scales and
aspect ratios. For each anchor, it predicts whether it contains an object
(object score) and refines the coordinates of the bounding box.
• The proposed regions from the RPN are fed into a ROI pooling layer to
extract fixed-sized feature maps for each region.
• ROI pooling allows different-sized regions to be scaled to the same size,
so they can be passed through fully connected layers.
• After ROI pooling, each region goes through a second-stage classifier to
predict the class label of the object within the region and further refine
the bounding box coordinates.

11 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

• Similar to YOLO, Faster R-CNN applies non-maximum suppression to

remove redundant bounding boxes and retain the most accurate
predictions.
Given an image of a building under construction, Faster R-CNN would first
propose candidate regions where objects like cranes, construction workers,
or materials might be present, and then classify and localize them with
bounding boxes.
Classification of worker safety in an image as done by R-CNN is shown in
figure 5.16.

Fig. 5.15. Object recognition by YOLO algorithm on an image

Before safety classification After safety classification

Fig. 5.16. Construction worker safety classification by R-CNN

12 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

Table 5.5. Intuitive explanation of optical flow related algorithms

Algorithm What it Does? How it Does?
Lucas-Kanade Algorithm The Lucas-Kanade algorithm is used to estimate optical flow,
which refers to the apparent motion of objects, surfaces, or
edges between two consecutive frames in a sequence of images
or a video. It assumes that the motion of objects between two
frames is small and relatively constant within a small
neighbourhood.
Lucas-Kanade is a local method, meaning it calculates the
optical flow in small patches of the image.
It works well for small, incremental motion but may not
perform well when objects undergo large motion between
frames.

In a video showing cars moving on a road, the Lucas-Kanade

algorithm can estimate the direction and speed of each car’s
motion by calculating the optical flow vectors for each car’s
pixels between frames.

Horn-Schunck Algorithm: Horn-Schunck is another method used to estimate optical flow,

like Lucas-Kanade, but it takes a global approach. Instead of
focusing on small patches, it computes a smooth flow field
across the entire image. It ensures that the estimated flow is
smooth by adding a regularization term..
In a construction site video, Horn-Schunck can compute the
motion of cranes, workers, and moving vehicles. The
algorithm will produce a smooth flow field, showing how
different parts of the scene move.
5.5 Image Classification
Image classification involves assigning labels to entire images based on the objects or patterns
they contain. Convolutional Neural Networks (CNNs) are the leading approach for such tasks.
A CNN begins by processing the input image, converting it into a feature map through multiple
convolutional and pooling layers. The convolutional layer applies a series of filters to the
image, with each filter generating a feature map that emphasizes a specific characteristic of the
input. Pooling layers then downsample these feature maps, reducing their size while preserving
essential information. The resulting feature maps undergo further convolutional and pooling
layers, allowing the network to learn increasingly complex features. Ultimately, the network
outputs a predicted class label or a probability score for each possible class, depending on the
taskThe layers of a Convolutional Neural Network (CNN) can be broadly classified into the
following categories:

13 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

Convolutional Layer: The convolutional layer is tasked with extracting features from the input
image. It applies a filter or kernel to the image, performing a convolution operation that
identifies and extracts specific features.
Pooling Layer: The pooling layer reduces the spatial dimensions of the feature maps generated
by the convolutional layer. Through down sampling, it minimizes the size of the feature maps,
lowering computational demands.
Activation Layer: The activation layer applies a non-linear activation function, like ReLU, to
the output of the pooling layer. This introduces non-linearity into the model, enabling it to learn
more complex patterns from the input data.
Fully Connected Layer: This layer connects all the neurons from the previous layer to all
neurons in the next, forming a traditional neural network. It combines the features learned in
earlier layers to make a prediction.
Normalization Layer: The normalization layer performs operations, such as batch or layer
normalization, to stabilize the activations within each layer, ensuring better conditioning and
reducing the risk of overfitting.
Dropout Layer: To prevent overfitting, the dropout layer randomly deactivates neurons during
training. This encourages the model to generalize well to new, unseen data, rather than simply
memorizing the training set.
Dense Layer: After features have been extracted by the convolutional and pooling layers, the
dense layer combines them to make a final prediction. Typically, the last layer in a CNN, it
receives flattened activations from previous layers, performs a weighted sum, and applies an
activation function to generate the final output. Figure 5.17 provides a schematic view of the
CNN.
For example, CNNs could be used for classifying types of construction materials (such as
identifying different types of concrete or bricks from images) or classifying land-use types
from aerial images for environmental impact assessments.

Fig. 5.17. A schematic view of CNN

14 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

5.6 Image Registration and Recognition

Image registration refers to the process of aligning two or more images of the same scene,
captured at different times, from varying angles, or using different sensors, so that they
overlap precisely. This is crucial in areas such as medical imaging, satellite image analysis,
and computer vision, where comparing or merging images from multiple sources is
necessary.

When images are taken under different conditions—such as at different times, angles, or with
different sensors—they may not be perfectly aligned. Image registration adjusts them to a
common coordinate system, ensuring accurate comparison and integration of data across
multiple images.

On the other hand, image recognition is the task of identifying objects, people, scenes, or
features within an image. It classifies images into pre-defined categories based on the patterns
and features present.

With a broad range of applications, image recognition is used in facial recognition, object
detection, autonomous vehicles, and medical imaging. It essentially allows machines to "see"
and interpret images similarly to how humans do.

5. CV applications in Civil Engineering

Computer vision, as a simplified imitation of the human visual system, is designed to provide
detailed information about both static and dynamic scenes. With the increasing affordability of
visual sensors and the availability of advanced visual systems, computer vision is becoming
integrated into various aspects of daily life. In the field of civil engineering, it is applied in
areas such as on-site construction, damage detection, data anomaly identification, structural
deformation measurement, safety and health monitoring, vibration analysis, and infrastructure
inspection. These applications are crucial for ensuring material integrity, serviceability, and
long-term sustainability. The following sections elaborate on these specific applications.
6.1 Construction Activity Monitoring
Collecting visual data about construction activities such as resource movements, movements
of personal, and equipment tracking will aid the construction managers in taking suitable
decisions and corrective actions upon noticing deviations. Algorithms have been developed to
detect the actions of excavators and earth-moving equipment by analyzing visual data extracted from
images, enabling the assessment of construction activities. Regular visual inspections around the
perimeter of concrete columns help identify discrepancies between the actual construction (as-built)
and the planned design (as-planned).

In another application, stereo cameras were employed to capture the movements of

construction workers, allowing for the evaluation of their impact on safety management. To
facilitate tracking on a construction site, entities are organized into clusters, with segmentation
algorithms used to monitor changes within these clusters. Additionally, a 3D transformation of
a series of 2D images captured at the construction site was utilized to oversee construction
progress. This process involves creating a 3D model of the site from the 2D images and
comparing this as-built model with the as-planned Building Information Modeling (BIM)
model to track progress effectively. . One such monitoring is displayed in figure 5.18.

15 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

Fig 5.18. Some use cases of construction activity monitoring

6.2 Construction Safety Monitoring

It is imperative and very much essential to monitor possible hazards at the construction site.
Quite a good number of applications of CV has taken place and reported thus far. The images
and videos obtained through automated site observation will complement in a big way in
relation to manual observations to circumvent risks. Pragmatically, the safety at construction
site seems to be the sought-after topic for researchers in this area ever since research
publications came through. A general framework that embeds CV related algorithms has served
as the mode for safety enhancement on site and also for health monitoring of workers in real-
time. The system also had a warning system that was planned to prevent collisions between
human workers and heavy equipment. In another application, 3D point cloud data is used to
design automatic regulation compliance system that could address safety issues related to
scaffold work related issues. To ensure worker’s safety and also to visualize and monitor
dynamic construction sites, an AI based tracking and detection system has been developed.
Conjoint application of computer vison and pattern recognition is proposed to detect unsafe
practices or behaviors by human workers has been proposed. This application used both spatial
and temporal information and modelling has been done using deep learning techniques.
Recurrent neural nets conjointly with CNN are facilitating architectures facilitating long term
memory. Such frame works have shown accurate detection of motion in unsafe operations such
as ladder climbing. Among recent AI algorithms CNN has found relatively numerous
applications in assessment of labor and health of workers also in tracking more complex
individual risk assessment. Some cases of safety monitoring using CV are captured in figure
5.19.

16 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

Fig. 5.19. Some use cases of CV in safety monitoring

6.3 Monitoring and Inspection

It is a well-known fact that the manual monitoring and inspection for the assessment of the
condition of the infrastructure or any other civil engineering related endeavor is arduous,
expensive and may turn out to be dangerous. It is expected that in the near future, it will be
replaced by the frameworks or monitoring systems that are CV based The applications are
wide, it may range from crack detection in concrete to changes in the structural configuration.
CV based algorithms that can facilitate inspection and real time monitoring are researched .
A descriptor and detector based on what is termed as speeded up robust features has been
developed. This system enables construction engineer to detect points of interest on the site
based on certain predefined parameters. The application is the basis that has furthered
researchers to evaluate the descriptors objects in the field that pertains to construction activity.
With such a system with built-in algorithms, on site personnel can use cell phones, to update
the information related to project and interpret certain construction processes happening at the
site.
6.4 Surveying and Modelling
CV based systems as a visual data acquisition technique is gaining importance as an instrument
for modelling objects in as-is-where-is status particularly for surveying applications. In this
regard, Building Information Models (BIM) deserve a special mention. BIM is in place as a de
facto standard in the engineering, architecture, as well as construction industry for storing the
information on assets. Generation of asset information automatically by BIM is expected to
create a remarkable impact on the stakeholders of construction industry. Such systems are
becoming a readily available instrument for modeling of assets and applications related to
surveying. A platform that is very unique in terms of data acquisition from the point clouds of

17 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

infrastructure using UAVs has been developed and reported. The work has elaborated the
modalities of using UAV systems for collecting data and also evaluated its supremacy over
traditional methods of surveying.
The system has been evaluated for its performance in sites where excavation and earthmoving
equipment are predominant activities. Photogrammetric technique is largely implemented for
surveying, and it is popular too. A framework that can assess structural condition of the bridges
and to support the maintenance of the same as per the priory plan, this framework uses UAV
data. Surveying methods have been developed to identify the materials of construction and also
to monitor the progress of the construction work.
6.5 Deformation and Vibration Monitoring
CV techniques have found monitoring deformations in structures. In one such application, the
deformation of the soil field surrounding the pile foundation has been measured. The concept
of optical extensometer has been utilized to measure the average strain of a deformed pile
foundation under horizontal load. An improved feature point tracking algorithm has been
implemented for the purpose. The measurements were found to be precise when compared with
conventional measurements. A CV based system for displacement measurement has been
developed and reported. This system is capable of dealing with illumination and fog related
issues. The proposed system leverages on high-resolution imaging by including contextual,
spatial and temporal aspects. Series of experiments of measurements were done on two –span
and three-lane bridges. The system has proved to be efficient in overcoming the fog and
illumination related issues. In a realistic implementation, a high-speed digital camera is used
to capture high amplitude vibrations of buildings in terms of inter-story drifts and the connected
personal computer captured the concurrent signals.
The presence of faults was detected by analyzing signals. Roving camera was used to capture
the response of a model bridge subjected to live loads. Indicators were developed to assess the
global displacement magnitude. The results are said to be encouraging. The significant take
away from this method is avoidance of cumbersome accelerometer cluster setup with
synchronized cameras.
6.6. Infrastructure inspection and monitoring
Infrastructure inspection involves two phases, deploying unmanned arial vehicles for image
data acquisition and processing the data to enable inspection. With the rapid growth of drone
industry, UAVs are becoming viable option for remote collection of data and they are longer
the thing of future. There are three primary approaches to damage detection from visual data:
heuristic-based methods, deep learning techniques, and change detection.
Traditionally, infrastructure monitoring involved measuring physical parameters such as strain,
displacement, acceleration, and natural frequency, achieved by placing wireless sensors at key
locations. However, sensor installation and maintenance can be costly. With the advancement
of computer vision (CV), non-contact methods for capturing data have become increasingly
popular. Algorithms can now establish optical flow, which identifies pixel displacement
between two images. Using this method, displacements and resulting strains can be determined
by comparing images of a critical location taken at different times. This approach, known as
Digital Image Correlation (DIC), has been applied to measure the deflection of a highway
bridge deck under static loading from 32-ton trucks. Numerous applications, particularly in

18 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

measuring deformations in bridge components, are well-documented in the literature. Another

significant use of DIC is in mapping cracks in areas experiencing high strain magnitudes.
Researchers have also employed this technique to map cracks in 3D images of reinforced
concrete beams and masonry walls subjected to dynamic loading.A. Detection of cracks
In most of the earlier CV based detection of cracks in concrete structures, heuristic image filters
such as edge and morphology based filters were used to determine the extent and thickness of
cracks. To classify the cracks of different thickness, ANNs and SVMs were used. A method is
developed to quantify the crack width first identifying the centre and then determining the edge
distances. Efforts have also been made to develop methods to elicit the nature and properties
of the cracks. In most of the cases, median filter is used to delineate the crack into edges and
skeletal. The morphological features were fed into neural networks and support vector
machines, to identify cracks of different thicknesses. One such outcome of an application is
presented in Fig. 5.19. The first column of this figure indicates original images; the second
column represents the crack detection by bottom hat method, while the third column is by
canny edge detection, and the last column. The methods so developed have found applications
over different kinds of structures including bridges, linings of tunnels and the cracks that
developed in the aftermath of earthquakes.
B. Detection of Spalling
Spalling, a surface defect, can compromise the durability and structural integrity of concrete,
while also negatively impacting its appearance. Traditionally, spalling is detected through
human visual inspection. Unlike other applications of computer vision (CV), substantial
research has yet to be conducted on detecting surface defects such as spalling. To address this,
an integrated model has been developed using hybrid algorithms in conjunction with regression
analysis. In this model, images are pre-processed using hybrid algorithms to reduce noise, and
critical features related to spalling are extracted from 3D images. Additionally, the spalling
depth is measured through regression analysis alongside image processing techniques.
In an attempt, an integrated model was validated using 75 images of spalled surfaces. An
orthogonal transformation method was applied to 3D images to quantify spalling on concrete
bridges, achieving 85% accuracy in detecting spalling. However, the method struggled when
spalling was accompanied by concurrent cracks. In another application, 3D images of spalled
columns were processed using segmentation, morphological operations, and template
matching, which proved to be an effective approach.
C. Detection of fatigue cracks and corrosion in Steel
Repeated impact loads from moving wheel traffic often lead to the formation of cracks in bridge
decks, particularly in steel bridges. Despite this, computer vision (CV) applications for
detecting such cracks remain limited. In one study, fatigue cracks were manually induced
around bolt holes, and the 3D images of these cracked areas were analyzed using filtering,
region of interest (ROI) segmentation, and object identification techniques. In another study,
images of cracked navigational meter gates and internal components of nuclear reactors were
analyzed using the region-growing method.
There is a notable body of research focused on developing CV-based approaches for detecting
corrosion. A comprehensive review of CV-assisted corrosion detection is available in the
literature. Typically, textural, spectral, and color information extracted from images has been

19 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

employed to assess the extent of corrosion. A widely used method combines wavelets with
principal component analysis (PCA). Additionally, robotics and smartphone-based
maintenance systems have been developed to detect corrosion through image analysis.
D. Detection of Cracks in Pavements
sizable number of CV based applications pertaining to detection of defects in pavements have
been reported. A comprehensive review on defect detection in asphaltic pavements is found in
reference [44]. In most of the applications, heuristic feature extraction, binary pattern-based
algorithm, and Gabor filtering methods have been used to detect the defects [45-48].
6.7 Applications in other Allied Areas
There is a wide array of allied applications of computer vision (CV), though such applications
are rarely discussed in the literature. These applications are often unique and discrete in nature.
For example, to determine the load rating of bridges, a cost-effective, practical, and novel field-
testing system utilizing portable cameras has been used to assess lateral live load distribution
factors in highway bridges. This approach has proven advantageous for two main reasons:
unlike conventional methods, traffic does not need to be halted, and it provides accurate values
for load distribution factors.
Additionally, a video image feature processing system has been proposed for measuring the
dynamic response of cables and determining cable tension. The algorithm developed accounts
for all types of loading conditions, and the feature-based image processing is integrated for
cable tension monitoring and dynamic analysis. In another related application, a non-contact,
vision-based system has been explored for similar purposes.

20 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

Fig 5.19. Crack detection

A non-contact, video-based sensor has been developed to measure cable tension forces,
eliminating the need for traditional sensors and the associated risks and challenges of
installation. This system was applied to measure tensile forces in the cables supporting a roof
structure, and the results showed good agreement between the forces measured by the vision-
based sensors and those recorded by load cells. This innovation offers a low-cost and
convenient solution for both long- and short-term monitoring of cable-supported structures.
In another interesting application of computer vision, a system was developed to determine the
proximity between workers and construction equipment. Built on Convolutional Neural
Network (CNN)-based image analysis, the system processes closed-circuit video footage to
ensure safety on construction sites.

6. Future Scope and Perspectives

Before delving into the future scope of computer vision (CV) applications in civil engineering,
it is essential to reflect on recent developments. A comprehensive review of the literature

21 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

reveals substantial advancements in CV applications within civil engineering. Notable progress

has been made in allied fields such as construction, operations and management, resource
optimization, monitoring, inspection, activity tracking, smart health monitoring, and distress
identification. However, this section outlines the extended agenda for expanding CV
applications across all aspects of civil engineering. The sub-domains where future CV
applications hold promise are as follows
• With Industry 4.0 in the vogue, for the researchers aiming to develop smart construction
sites, there is a huge scope to deal with data provided by visual systems particularly
concerning safety of workers, resource management, infrastructure management and
smart health monitoring to possibly develop a framework for safe, efficient and
networked construction management systems.
• Integration of the currently established visual data acquisition systems into IoT has a
huge scope in terms of improved cost benefit ratio apart from furthering the network of
researchers, developers and practicing engineers.
• Availability of knowledge from smart construction sites, will go a long way in
developing sophisticated computer vision algorithms that can assess and systems that
are able to provide feedback on real time basis.
• With huge generation of data from computer vision systems,the storage will definitely
pose a challenge in this direction, researchers will have quite a good number of open-
ended questions and opportunities to develop ontology models to decipher and to
manage knowledge acquired through CV based systems.
• All said and done, the performance and quality standards in construction industry is still
being placed at a lower score particularly with reference to material and infrastructure
maintenance,life expectancy, quality of construction, and monitoring. A rethinking to
address the related issues has a huge scope for improvement.
• Construction industry in general will encompass voluminous materials of different
kind, different shapes of built-in components, and used products, research on
automation in managing these has not yet started. Therefore, it is envisaged that the
development of automatic quality inspection for constructed components, vison-based
quality checking for extruded products will pick up momentum in coming days.
• From the available literature though sparse, there seems to be surge of interest in
adopting CV in off-site construction (OSC) in the industry. However, CV based
applications seem to elude this area. This will surely provide several open problems for
prospective researchers.
• There are many challenges and technical bottlenecks in the area of CV based automated
vibration monitoring systems. The crux of the issue here is transformation of features
obtained by vision-based measurements in to interpretable information to assist in
making decisions. This area is still open for exploration.
• Computer vision blended with deep learning algorithms (or any other AI related
algorithm for that matter) can hardly replicate a trained civil engineer at site. Also, the
implications of flaws, and various kinds of damage can be quickly realized by a human
expert.
• Research in the area of precise damage detection, recognition of the component, and
interpretation of the damage for making local and overall assessment of the damage is
not found in the literature. This is yet another area for exploration.

22 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

• As the structures in operation will undergo deformations (displacements and associated

strains) of smaller magnitude, the accessibility of the component of the structure is often
limited.
• This prohibits optimal camera locations for taking measurements of high quality. In this
direction efforts are needed to develop both software and compatible hardware to
facilitate CVbased strain and displacement measurement systems.
• Civil infrastructure by nature composes of mixed environment of various sized (small
– large) components. This makes selection of distinctive features of the component
challenging therefore, implementers of CV systems will end up in feature selection
case-by-case, thus missing overall assessment.
• It seems that studies on use of CV for inspections that are product centric in terms of
detection of flaws are very limited. In this direction, researchers may work hand-in-
hand with industry with a greater focus to attract funding by stake holders.
From the foregone observations, it may be said that CV is certainly a natural choice for almost
all types of civil engineering endeavours. However, image in itself will provide rich textural,
and spatial data. Deriving information that is measurable and actionable is really challenging.
Equally enough, the researchers are striving hard and have achieved a degree of success in
showcasing the potential of CV related techniques ranging from optical flow to deep learning.

Module End Questions

1. Present your understanding on Computer Vision. What are phases of CV? Illustarte with
examples.
2. Give an abstract view on CV and its applications in civil engineering.
3. Elaborate on image acquisition methods. Also give a list of tools for the same.
4. Discuss on steps of image processing after the acquisition of images.
5. Write brief notes on :
i. Image segmentation and related algorithms.
ii. Image registration
iii. Optical flow
iv. Object detection
v. Image classification
vi. CNN and its applications
vii. YOLO
6. Present on various real world applications of CV in civil engineering.

References
1) D.A. , J. Ponce, Computer vision: a modern approach, Prentice Hall Professional Technical Reference,
Upper Saddle River, 2002.
2) David Lowe, The computer Vision Industry, available: https://www.cs.ubc.ca/_lowe/vision.html,
accessed on 11 September 22.
3) J. Yang,(2015) Construction performance monitoring via still images, time-lapse photos, and video
streams: now, tomorrow, and the future. Adv Eng Inform,2015; 29(2) :211-224,
https://doi.org/10.1016/j.aei.2015.01.011.

23 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

4) K.K. Han, M. Goparvar Fard,(2017) Potential of big visual data and building information modelling for
construction performance analytics: an exploratory study, Autom Constr 73 (2017) 184–198,
https://doi.org/10.1016/j.autcon.2016.11.004.
5) D. Wang, F. Dai, X. Ning, Risk assessment of work-related musculoskeletal disorders in construction:
state-of-the-art review, J Constr Eng Manag 141 (6) (2015) 040150086,
https://doi.org/10.1061/(ASCE)co.1943-7862.0000979.
6) J. Seo, S. Han, S. Lee, H. Kim, Computer vision techniques for construction safety and health monitoring,
Adv Eng Inform 29 (2) (2015) 239–251, https://doi.org/10.1016/j.aei.2015.02.001.
7) J. Teizer, Status quo and open challenges in vision-based sensing and tracking of temporary resources on
infrastructure construction sites, Adv Eng Inform 29 (2) (2015) 225–238,
https://doi.org/10.1016/j.aei.2015.03.006.
8) M.A. Jayaram. (2023) Computer vision applications in construction material and structural health
monitoring: A scoping Review, Materials Today: Proceedings,2023,ISSN 2214-
7853,https://doi.org/10.1016/j.matpr.2023.06.031.
9) Zeliski, R. (2010). Computer Vision: Algorithms and Applications. Springer.
10) Hartley, R., & Zisserman, A. (2003). Multiple View Geometry in Computer Vision. Cambridge University
Press.
11) Ren, S., He, K., Girshick, R., & Sun, J. (2015). "Faster R-CNN: Towards Real-Time Object Detection
with Region Proposal Networks." IEEE Transactions on Pattern Analysis and Machine Intelligence,
39(6), 1137-1149.
12) S. Han, S. Lee, A vision-based motion capture and recognition framework for behavior-based safety
management, Autom. Constr. 35 (2013) 131–141,https://doi.org/10.1016/j.autcon.2013.05.001.
13) N. Silberman, D. Hoiem, P. Kohli, R. Fergus, Indoor segmentation and support inference from RGBD
images, in: A. Fitzgibbon, S. Lazebnik, P. Perona, Y. Sato,C. Schmid (Eds.), Computer Vision – ECCV
2012. DOI:10.1007/978-3-642-33715-4_54.
14) W.R. Abdulmajeed, R.Z. Mansoor, Implementing kinect sensor for building 3D maps of indoor
environments, International Journal of Computer Applications , 2014;86(8):1-8.Doi:10-5120/15005-
3182.
15) S. McMahon, N. Sünderhauf, B. Upcroft, M. Milford, Multimodal trip hazard affordance detection on
construction sites, IEEE Robotics and Automation Letters 3 (1) (2017) 1–7. 10.48550/arXiv.1706.06718.
16) H. Son, H. Seong, H. Choi, C. Kim, Real-time vision-based warning system for prevention of collisions
between workers and heavy equipment, J. Comput.Civ. Eng 33 (5) (2019) 1–10,
https://doi.org/10.1061/(asce)cp.1943-5487.0000845.
17) Q. Wang, Automatic checks from 3D point cloud data for safety regulation compliance for scaffold work
platforms, Autom. Constr 104 (2019) 38–51,https://doi.org/10.1016/j.autcon.2019.04.008.
18) K.K. Han, M. Golparvar-Fard, Appearance-based material classification for monitoring of operation-
level construction progress using 4D BIM and site photologs, Automation in Construction 53 (2015) 44–
57, https://doi.org/10.1016/j.autcon.2015.02.007.
19) X. Luo, H. Li, H. Wang, Z. Wu, F. Dai, D. Cao, Vision-based detection and visualization of dynamic
workspaces, Autom. Constr 104 (2019) 1–13,https://doi.org/10.1016/j.autcon.2019.04.001.
20) L. Ding, W. Fang, H. Luo, P.E. Love, B. Zhong, X. Ouyang, A deep hybrid learning model to detect
unsafe behavior: integrating convolution neural networks and long short-term memory, Autom. Constr
86 (2018) 118–124, https://doi.org/10.1016/J.AUTCON.2017.11.002.
21) W. Fang, L. Ding, H. Luo, P.E. Love, Falls from heights: a computer vision based approach for safety
harness detection, Autom. Constr. 91 (2018) 53–61,https://doi.org/10.1016/J.AUTCON.2018.02.018.
22) X. Luo, H. Li, D. Cao, Y. Yu, X. Yang, T. Huang, Towards efficient and objective work sampling:
recognizing workers’ activities in site surveillance videos with two stream convolutional networks,
Autom. Constr. 94 (2018) 360–370,https://doi.org/10.1016/J.AUTCON.2018.07.011.
23) J. Chen, Y. Fang, Y.K. Cho, Performance evaluation of 3D descriptors for object recognition in
construction applications, Autom. Constr. 86 (2018) 44–
52,https://doi.org/10.1016/j.autcon.2017.10.033.
24) H. Bae, M. Golparvar-Fard, J. White, High-precision vision-based mobile augmented reality system for
context-aware architectural, engineering, construction and facility management (AEC/FM) applications,
Visualization in Engineering 1 (1) (2013) 3–10, https://doi.org/10.1186/2213-7459-1-3.
25) V. Pa˘tra˘ucean, I. Armeni, M. Nahangi, J. Yeung, I. Brilakis, C. Haas, State of research in automatic as-
built modelling, Adv. Eng. Inform. 29 (2) (2015) 162–171, https://doi.org/10.1016/j.aei.2015.01.001.

24 CV and Its applications in Civil Engineering

AI and Applications in Civil Engineering

26) Y. Ham, K.K. Han, J.J. Lin, M. Golparvar-Fard, Visual monitoring of civil infrastructure systems via
camera-equipped unmanned aerial vehicles (UAVs): A review of related works, Visualization in
Engineering. 4 (1) (2016)1–12.
27) S. Siebert, J. Teizer, Mobile 3D mapping for surveying earthwork projects using an Unmanned
AerialVehicle (UAV) system, Autom. Constr. 41 (2014) 1–14,
https://doi.org/10.22260/ISARC2013/0154.
28) S. Kang, M.W. Park, W. Suh, Feasibility study of the unmanned-aerial-vehicle radiofrequency
identification system for localizing construction materials on large-scale open sites, Sensors and
Materials 31 (5) (2019) 1449–1465, https://doi.org/10.18494/SAM.2019.2266.
29) Kun Zhou1, Linhua Chen1, Shanshan Yu, Vision based deformation measurement Pile soil testing, In
proceedings of MATEC Web of Conferences, 2019;275:1-9. DOI:10.1051/matecconf/201927503009.
30) Chuan-Zhi Dong , Ozan Celik, F. Necati Catbas , Eugene OBrien , Su Taylor, A Robust Vision-Based
Method for Displacement Measurement under Adverse Environmental Factors Using Spatio-Temporal
Context Learning and Taylor Approximation, Sensors, 2019;19:1-22. doi: 10.3390/s19143197.
31) Wei Wang, Yu Shao, Building Vibration Monitoring Based on Digital Optical Cameras, Journal of Vibro
Engineering, 23(6) (2021) 1383-1394. https://doi.org/10.21595/jve.2021.21999.
32) D. Lydon, M. Lydon, R. Kromanis, ChuanZhi Dong, S.u. Necati Catbas, Taylor,, Bridge Damage
Detection Approach Using a Roving Camera Technique, Sensors 21 (2021) 1–22,
https://doi.org/10.3390/s21041246.
33) B.F. Spencer Jr, Vedhus Hoskere, Yasutaka Narazak, Advances in Computer Vision-Based Civil
Infrastructure Inspection and Monitoring, Engineering 5(2019) 199–222,
https://doi.org/10.1016/j.eng.2018.11.030.
34) M.R. Jahanshahi, J.S. Kelly, S.F. Masri, G.S. Sukhatme, A survey and evaluation of promising
approaches for automatic image-based defect detection of bridge structures, Struct Infrastruct Eng 5 (6)
(2009) 455–486, https://doi.org/10.1080/15732470801945930.
35) T. Dawood, Z. Zhu, T. Zayed, Machine vision-based model for spalling detection and quantification in
subway networks, Automation in Construction 81 (2017) 149–160,
https://doi.org/10.1016/j.autcon.2017.06.008.
36) C.M. Yeum, S.J. Dyke, Vision-based automated crack detection for bridge inspection, Comput Civ
Infrastruct Eng. 30 (10) (2015) 759–770, https://doi.org/10.1111/mice.12141.
37) M.R. Jahanshahi, F.C. Chen, C. Joffe, S.F. Masri, Vision-based quantitative assessment of microcracks
on reactor internal components of nuclear power plants, Struct Infrastruct Eng. 13 (8) (2017) 1–8.
38) S.K. Ahuja, M.K. Shukla, A survey of computer vision-based corrosion detection approaches. In:
Satapathy S, Joshi A, editors. Information and communication technology for intelligent systems (ICTIS
2017), 2018; 2:Springer.55-63.
39) S. Ghanta, T. Karp, S. Lee, Wavelet domain detection of rust in steel bridge images. In: Proceedings of
2011 IEEE International Conference on Acoustics, Speech and Signal Processing; 2011 May 22–27;
Prague, Czech Republic.Piscataway: IEEE.1033-36.
40) M.R. Jahanshahi, S.F. Masri, Parametric performance evaluation of wavelet based corrosion detection
algorithms for condition assessment of civil infrastructure systems, J Comput Civ Eng. 27 (3) (2013) 45–
57, https://doi. org/10.1061/(ASCE)CP.1943-5487.0000225.
41) S. Jang, H. Jo, S. Cho, K. Mechitov, J.A. Rice, S.H. Sim, et al., Structural health monitoring of a cable-
stayed bridge using smart sensor technology: deployment and evaluation, Smart Struct Syst. 6 (5) (2010)
439–459,https://doi.org/10.1117/12.852272.
42) J.A. Rice, K. Mechitov, S.H. Sim, T. Nagayama, S. Jang, R. Kim, Flexible smart sensor framework for
autonomous structural health monitoring, Smart Struct, Syst 6 (6) (2010) 423–438.
10.12989/sss.2010.6.5_6.423.
43) Y.F. Liu, S.J. Cho, B.F. Spencer, Jr., J.S. Fan, Concrete crack assessment using digital image processing
and 3D scene reconstruction, J Comput Civ Eng, 2016;30(1):DOI:10.1061/(ASCE)CP.1943-
5487.0000446.
44) C. Koch, I. Brilakis, Pothole detection in asphalt pavement images, Adv Eng Inform. 25 (3) (2011) 507–
515, https://doi.org/10.1016/j.aei.2011.01.002.
45) M. Salman, S. Mathavan, K. Kamal, M. Rahman, Pavement crack detection using the Gabor filter. In:
Proceedings of the 16th International IEEE Conference on Intelligent Transportation Systems, Oct 6–9,
2013, Hague, Netherlands.2039–44.

25 CV and Its applications in Civil Engineering

 AI and Applications in Civil Engineering

46) Y. Hu, C. Zhao, A novel LBP based methods for pavement crack detection, J Pattern Recognition Res. 5
(1) (2010) 140–147.
47) Q. Zou, Y. Cao, Q. Li, Q. Mao, S. Wang, Crack Tree: automatic crack detection from pavement images,
Pattern Recognit Lett 33 (3) (2012) 227–238, https://doi.org/10.1016/j.patrec.2011.11.004.
48) L. Li, L. Sun, S. Tan, G. Ning, An efficient way in image preprocessing for pavement crack images. In:
Proceedings of the 12th COTA International Conference of Transportation Professionals, Aug 3–6;
Beijing, China. 2012,3095-3103.
49) S. Yoneyama, A. Kitagawa, S. Iwata, K. Tani, H. Kikuta, Bridge deflection measurement using digital
image correlation, Exp Tech l31(1) (2007) 34-40.
50) S. Yoneyama, H. Ueda, Bridge deflection measurement using digital image correlation with camera
movement correction, Mater Trans 53 (2) (2012),285–290, https://doi.org/10.2320/MATERTRANS.I-
M2011843.
51) M.N. Helfrick, C. Niezrecki, P. Avitabile, T. Schmidt, 3D digital image correlation methods for full-field
vibration measurement, Mech Syst Signal Process 25 (3) (2011) 917–927,
https://doi.org/10.1016/j.ymssp.2010.08.013.
52) R. Ghorbani, F. Matta, M.A. Sutton, Full-field deformation measurement and crack mapping on confined
masonry walls using digital image correlation,Exp Mech. 55 (1) (2015) 227–243,
https://doi.org/10.1007/s11340-014-9906-y.
53) Chuan Zhi Dong, Selcuk Bas, F. Necati Catbas, Portable monitoring approach using cameras and
computer vision for bridge load rating in smart cities,Journal of Civil Structural Health Monitoring,10
(2020) 1001-1021.https://doi.org/10.1007/s13349-020-00431-2.
54) C. Chu, F. Ghrib, S. Cheng, Cable tension monitoring through feature-based video image Processing,
Journal of Civil Structural Health Monitoring 11 (2021) 69–84, https://doi.org/10.1007/s13349-020-
00438-9.
55) D. Feng, T. Scarangello, M.Q. Feng, Q.i. Yea, Cable tension force estimate using novel noncontact
vision-based sensor, Measurement 99 (2017) 44–52. 10.1016%2Fj.measurement.2016.12.020.
56) Y.-S. Shin, J. Kim, A Vision-Based Collision Monitoring System for Proximity of Construction Workers
to Trucks Enhanced by Posture-Dependent Perception and Truck Bodies’ Occupied Space, Sustainability
14 (2022) 1–13, https://doi. org/10.3390/su14137934.

26 CV and Its applications in Civil Engineering

View publication stats