International Journal of Scientific Research in Engineering and Management (IJSREM)

Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

AI-Enhanced Building Safety:

Real-Time Crack Detection with YOLO
Mrs.R.Aruna 1, M.Umar Farook 2, P.Vignesh 3, S.Santhosh Kumar 4
1
Assistant Professor, Department of IT , Kongunadu College of Engineering and Technology
2
Student, Department of IT , Kongunadu College of Engineering and Technology
3
Student, Department of IT , Kongunadu College of Engineering and Technology
4
Student, Department of IT , Kongunadu College of Engineering and Technology
--------------------------------------------------------------------- ------------------------------------------------------------------------
Abstract - - In an increasing demand for safe and more OpenCV for image processing. It improves overall building
reliable infrastructure, early detection of structural defects in safety by offering continuous monitoring through the integration
buildings is now considered to prevent a catastrophic failure. This of object detection, cloud storage, and automated reporting.
paper discusses an AI-enhanced real-time crack detection system
using the YOLO algorithm, one of the deep learning algorithms 2. METHODOLOGY
which can detect an object from the given images or video feeds
in real time. The proposed system makes use of YOLO's object
detection capability to detect and classify cracks in images or Safety in buildings is essential, particularly in older buildings that
video feeds. This system aims to support the efficient and are subject to harsh weather conditions. Manual inspections are
proactive monitoring of structure health by building managers laborious and prone to mistakes. With the help of YOLO, this
and engineers. Delivering fast and accurate crack detection, this study presents an AI-powered crack detection system that
AI solution surpasses the traditional approaches that would provides real-time alerts via email, SMS, or apps for prompt
normally raise the costs of maintenance and minimize the overall action. YOLO improves maintenance and lowers repair costs by
safety level of a building. Implementation of this methodology achieving over 90% accuracy in real-time crack detection.
would help result in safer buildings and cost-efficient building
management.
2.1Related Works
Key Words:AI-enhanced crack detection ,YOLO algorithm [1] An AI driven crack identification technique for stone masonry
Real-time monitoring ,Structural health, Building safety.
that operates without labeled data was presented by Agrafiotis et
al. For historic structures where manual checks are difficult, this
is ideal.
1. INTRODUCTION
[2] Golding et al. achieved great accuracy in identifying concrete
Building safety is one of the significant concerns, especially in cracks by using deep learning. Their CNN-based model performs
aged structures and areas prone to extreme weather or seismic better than conventional techniques, resulting in quicker and
activity. Traditional monitoring methods such as manual more accurate inspections.
inspections are time-consuming and full of mistakes. This paper
[3] To detect pavement cracks, Hu and Zhao used Local Binary
presents an AI-powered real-time crack detection system using
Patterns (LBP). Their method is excellent at managing various
the YOLO deep learning model. The developed system
surface conditions and sunlight, opening the door for
automatically detects and classifies various types of cracks which
contemporary AI crack detection.
will provide building managers with aproactive solution towards
issues that may arise before they deteriorate, therefore reducing [4] To improve building façade defect identification, Interlando
repair costs and safety risks. Real-time alerts are integrated in the et al.'s study makes use of several deep learning models. Their
system to alert stakeholders by email, SMS, or application when method works very well for monitoring cities on a broad scale,
there is a crack found. Cost-effectiveness and scalability are which helps to make infrastructure management safer.
offered with the help of open-source tools like OpenCV for
image processing and TensorFlow or PyTorch for running [5] Medina et al. presented a hybrid approach for detecting road
YOLO. surface cracks that incorporates 2D and 3D images. This
integration improves accuracy and drastically lowers false
This combines object detection, cloud storage, and automated positives, making it extremely dependable for practical uses.
reporting to monitor the site continuously and making buildings
safer overall. When a crack is found, the system immediately [6]For accurate pavement distress identification, Mathavan et al.
notifies stakeholders by email, SMS, or app, allowing for timely examined 3D imaging methods as laser scanning and stereo
action. It guarantees affordability and scalability by utilizing vision. These methods are popular for monitoring infrastructure
open-source tools such as TensorFlow/PyTorch for YOLO and because of their great accuracy.

© 2025, IJSREM | www.ijsrem.com DOI: | Page 1

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

[7] To effectively detect road cracks, Oliveira and Correia created Fig 2.2.1.System Architecture
an automated method that combines edge recognition and
morphological picture processing. Their method offers a scalable A. Dataset Collection
and efficient way to evaluate transportation infrastructure.
Gathering a collection of crack photos is the initial stage in the
training phase. These datasets, which include tagged photos of
cracks and non-cracks, are taken from Kaggle or other
[8] In order to improve crack detection accuracy while preserving repositories. To increase the resilience of the model, the dataset
computing efficiency,Ren et al. integrated attention mechanisms should be varied, including pictures with various surfaces, angles,
into deep learning models. Their research yields encouraging and lighting conditions. The dataset is divided into training and
outcomes for practical implementations. validation sets, with the model being trained on the training set
and its performance assessed on the validation set. To increase
[9] Sarhadi et al. used an enhanced SWIN U-Net model with dataset variability, data augmentation methods such as flipping,
attention mechanisms to optimize the identification of concrete rotation, and contrast modifications can also be used.
cracks. Because of its great accuracy and dependability, this
approach is appropriate for monitoring concrete infrastructure on
a broad scale.
B. Model Construction Using YOLO Algorithm
[10] For fracture identification, Shu et al. presented an active
learning technique using a difficulty learning mechanism. By The entire image is processed in a single pass by the realtime
giving complicated samples priority, our method lowers object detection model known as the YOLO (You Only Look
annotation costs while enhancing model performance overall. Once) method. YOLO creates bounding boxes, separates an
image into grid cells, then categorizes items inside the
[11] A deep learning model created especially for identifying boxes.YOLO is trained to identify crack patterns and differentiate
cracks in marble surfaces was created by Vrochidou et al. them from non-crack areas in order to detect cracks. The number
Supporting robotic applications for automated resin application in of epochs, batch size, and learning rate are among the parameters
the marble industry is the goal of their work. that are set up for the model. Convolutional layers are used to
refine the network architecture in order to efficiently extract
[12] Varadharajan et al. talked about computer vision methods spatial data, guaranteeing excellent detection accuracy while
for road inspection that are powered by AI. Their model preserving processing speed in real time.
effectively detects pavement cracks and other road issues by
using image processing.

[13] A deep residual U-Net model for pavement fracture C. YAML File Generation
identification was presented by Yang et al. To improve
segmentation accuracy, their method makes use of deep feature Following training, the YOLO model creates a YAML (Yet
extraction and skip connections. Another Markup Language) file that serves as a structured
configuration file with all the necessary settings for deployment
[14] CNN-based deep learning models for concrete surface crack and testing. In order for the trained model to be effectively reused
detection were compared by Zadeh et al. With its great accuracy without retraining and be able to categorize fresh images with
in detecting structural flaws, their research demonstrates AI's high accuracy, this file is essential. By organizing important
expanding potential in infrastructure monitoring model metadata, it facilitates the trained model's integration into
other machine learning frameworks and practical applications.
The dataset configuration, which specifies paths for training,
2.2.System Architecture validation, and test datasets, is the first of several essential
elements included in the YAML file. The number of object
The software system automatically detects and analyzes
classes (nc) and their labels—typically ['crack'] in the context of
building flaws by combining cloud, AI, and computer vision
crack detection—are also specified. It also includes anchor box
technologies. Building managers can promptly resolve problems,
definitions, which are bounding boxes that have been designed to
stop additional damage, and lower maintenance
help detect cracks of various sizes and forms, hence increasing
the accuracy of detection. Hyperparameter storage, which
includes parameters like learning rate, batch size, and momentum
. that affect the model's stability and performance, is another
crucial area of the YAML file. The YAML file bridges the gap
between training and practical application by standardizing these
configurations, guaranteeing consistency, portability, and smooth
deployment of the trained model in diverse situations.

© 2025, IJSREM | www.ijsrem.com DOI: | Page 2

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

D. Accuracy Score Evaluation Classification Using Model File

Accuracy measures are used to assess the model's performance Noise Filtering: Uses bilateral, median, or Gaussian filtering to
after training. Predicted labels and ground truth labels are eliminate dust, shadows, and reflections. This increases the
compared to calculate the accuracy score. Model performance is accuracy of fracture detection, decreases false positives, and
measured using metrics like Intersection over Union (IoU), F1- improves image clarity .Key crack patterns, such as edges, lines,
score, precision, and recall. When a model has a high accuracy and textures, are identified by feature extraction. In order to
score, it can identify cracks with few false positives and false distinguish genuine cracks from dirt, scratches, or surface flaws,
negatives.In order to increase performance, hyperparameters are the YOLO model examines various areas.Classification: Detected
adjusted and more training data is added if the accuracy score is patterns are given a likelihood score by the trained YOLO model.
not adequate. During the testing phase, the accuracy score aids in The mechanism verifies the existence of a crack if the score rises
determining whether the model is prepared for deployment. above a predetermined level. Model weights that are stored (from
the YAML file) provide precise and effective classification of
fresh photos.

E. Input Image

A fresh image is supplied as input for crack identification during I. Crack Detection Output and Notification
the testing phase. This picture might have been captured by a System
camera or an industrial inspection tool. The image is used as the
raw data, which will go through a number of preprocessing stages The system logs the results for analysis after detecting a fracture
before the trained model analyzes it. The accuracy of detection is and marks it with bounding boxes. It is used to automate
greatly influenced by the quality of the input image. Clear inspections and reduces damage in buildings, bridges, and
contrast and high definition photos enhance the model's capacity roadways. The notification system provides location, severity,
to detect cracks. The final detection findings could be impacted and crack details to maintenance staff in real time, via software,
by any distortion, blurriness, or noise in the input image. emails, or SMS. This makes it possible to act quickly, avoiding
expensive repairs and guaranteeing safety.

F. Prepossessing 2.3.Workflow
The image is preprocessed to improve features and eliminate
extraneous parts before the model processes it. To increase
A. Image Input
detection accuracy, preprocessing entails scaling the image to fit
the input dimensions of the model and normalizing pixel values. An input image is first sent to the system for analysis. Usually,
Additional methods like sharpening filters and contrast cameras or drones are used to take this picture in order to inspect
augmentation can also be used. By following these procedures, the infrastructure. It ensures that the detection process begins
you may be sure that the image is in the best possible format for with appropriate and lucid visual data by acting as the basis for
feature extraction. Better detection results are achieved by further processing
minimizing the effects of illumination, texture, and background
noise fluctuations while assisting the model in concentrating on
crack-related variables through appropriate preprocessing.

G. Gray Scale Conversion

A crucial preprocessing step that converts a color image to a

black-and-white (grayscale) format is grayscale conversion. This
preserves crucial fracture structural information while lowering
computational complexity. The contrast between crack regions
and non-crack surfaces is improved by converting the image to
grayscale because cracks are usually depicted by changes in pixel
intensity rather than color. This improves the model's ability to
identify minute details in the picture. Grayscale conversion
makes sure that the texture and pattern of the cracks are the main
emphasis by removing extraneous color information.
Fig 2.3.1: Architecture diagram for Workflow

H. Noise Filtering ,Feature Extraction and

© 2025, IJSREM | www.ijsrem.com DOI: | Page 3

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

B. Filtering out noise G. Analysis & Action

Dust, shadows, and reflections are examples of undesired Maintenance personnel examine the detection data and take
components that can obstruct detection. By eliminating these corrective action as the last step. This contributes to the safety
distortions while maintaining crucial fracture information, noise and durability of infrastructure by preventing structural failures.
filtering techniques like Gaussian and median filtering improve
the image. Better accuracy in subsequent processing is ensured
by this phase. 2.4.Experimental Analysis

Fig.2.4.1 Input Preprocessing

C. Extraction of Features

Important visual elements are taken out of the picture, including

textures, edges, and lines. This procedure aids in differentiating Through step-by-step image analysis in above Figure 2.4.1,this
fractures from comparable formations such as surface AI-based method finds building fractures. Before enhancing
imperfections or scratches. The features that have been retrieved contrast to highlight cracks, the image is first converted to
are used as inputs for the classification model. grayscale to eliminate color distractions. It then uses binary
conversion, which isolates the cracks by making them look as
black lines on a white background, and segmentation. AI
evaluates these fissures to determine the extent of structural
D. The YOLO Model damage, allowing for prompt repair and early identification. By
preventing malfunctions and ensuring prompt repairs through
By examining the extracted features, the YOLO (You Only Look
real-time notifications, this increases building safety and makes
Once) model ascertains whether a crack is present. Using trained
structures stronger and more resilient.
data, it rapidly processes the image in real-time, identifying areas
as crack or non-crack.

E. Crack Identification

The technique uses bounding boxes to highlight any cracks that

are found. Maintenance personnel can find the problem more
easily thanks to this visual marker.

F. Notification System

When a crack is found, the notification system notifies users, Fig.2.4.2 Shadow Removal and Denoising
guaranteeing prompt action. Depending on the integration, Iteration
notifications may be provided by text message, email, or software
alerts. These notifications assist maintenance personnel prioritize
important cases by providing crucial information like the
The above Figure 2.4.2 demonstrates the methodical steps
location, severity, and kind of the fracture .The solution improves
involved in AI-based crack detection for building security.
efficiency, decreases human monitoring efforts, and stops
Although the first image shows the original grayscale surface, it
infrastructure faults before they worsen by providing realtime
is difficult to see the crack because of noise and shadows. The
notifications.
crack is accentuated and made more noticeable in the second
image by denoising and shadow removal. In the final image, the
AI precisely detects the crack by highlighting it in yellow. This
procedure facilitates real-time monitoring, permits prompt
repairs, and averts significant structural breakdowns.

© 2025, IJSREM | www.ijsrem.com DOI: | Page 4

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

For detecting structural fractures, the YOLO-based AI

crack detection system offers a quick, dependable, and realtime
solution. With little human labor, it guarantees great accuracy
through the use of deep learning and computer vision. YOLO is
scalable and effective when used with Python, PyTorch, and
OpenCV. This lowers inspection costs and permits early
maintenance to stop significant damage. To ensure prompt
response, the system has an enhanced alert mechanism that sends
administrators emails, SMS, and messages when cracks are
found. Engineers may make well-informed judgments with data
logging and real-time monitoring. A completely automated
infrastructure monitoring system may result from upcoming
improvements like cloud storage, predictive analytics, and IoT
Fig.2.4.3 Crack Detection
integration. By making building maintenance more proactive,
secure, and economical, this innovation ensures long-term
durability with little work.
The last phase of AI fracture identification on a concrete surface
is depicted in the above Figure 2.4.3. When the crack is indicated
in yellow, the AI has successfully located it. The procedure ACKNOWLEDGEMENT
entails eliminating shadows, cutting down on noise, and clearly
identifying the crack. This aids in prompt repairs, halting I sincerely thank my mentor, research team, and technical
experts for their invaluable guidance, feedback, and
additional harm and maintaining the safety of buildings.
dedication throughout this project. Special appreciation goes
to the developers of the YOLO algorithm and AI technologies.
I also

acknowledge my family's support in this endeavor.

REFERENCES

[1] Agrafiotis, P., Doulamis, A., & Georgopoulos, A.

"Unsupervised Crack Detection on Complex Stone Masonry
Surfaces." arXiv preprint arXiv:2303.17989, 2023.

Fig.2.4.4 Experimental Results [2] Golding, Vaughn Peter, et al. "Crack Detection in Concrete
Structures Using Deep Learning." Sustainability 14.13 (2022):
8117.
The above Figure 2.4.4 result is demonstrates how an AI system
gradually identifies cracks. The crack is successfully detected
after 500 attempts. The system monitors development and gets
[3] Hu, Y., and C. Zhao. "A Local Binary Pattern Based
better with time. Crack details such as confidence (0.6), intensity
Methods for Pavement Crack Detection." Journal of Pattern
(56136.0), area (1334.3), and diameter (267.3) are measured in
Recognition Research 5.1 (2010): 140–147.
the final result. This aids in preventative maintenance to ensure
the safety of buildings.

[4] Interlando, M., Pacifico, M. G., Novellino, A., & Pastore, V.

3. CONCLUSIONS P. "Ensembles of Deep Neural Networks for the Automatic
Detection of Building Facade Defects from Images." IEEE
Access, 2024.

© 2025, IJSREM | www.ijsrem.com DOI: | Page 5

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

[14] Zadeh, S. S., Birgani, S. A., Khorshidi, M., & Kooban, F.

"Concrete Surface Crack Detection with Convolutionalbased
[5] Medina, R., J. Llamas, E. Zalama, and J. Gomez Deep Learning Models." arXiv preprint arXiv:2401.07124, 2024.
GarciaBermejo. "Enhanced Automatic Detection of Road Surface
Cracks by Combining 2D/3D Image Processing Techniques." In
Proceedings of IEEE International Conference on Image [15] R. Sathya, V. C. Bharathi, S. Ananthi, K. Vaidehi and S.
Processing, 2014, pp. 778–782.
Sangeetha, "Intelligent Home Surveillance System using Convolution
Neural Network Algorithms," 2023 4th International Conference on
Electronics and Sustainable Communication Systems (ICESC),
Coimbatore, India, 2023, pp. 1683-1688, doi:
[6] Mathavan, S., K. Kamal, and M. Rahman. "A Review of
10.1109/ICESC57686.2023.10193402.
Three-Dimensional Imaging Technologies for Pavement Distress
Detection and Measurements." IEEE Transactions on Intelligent
Transportation Systems 16.5 (2015).
[16] R. Sathya, M. Mythili, S. Ananthi, R. Asitha, V. N. Vardhini and
M. Shivaani, "Intelligent Video Surveillance System for Real Time
Effective Human Action Recognition using Deep Learning
[7] Oliveira, H., and P. L. Correia. "Automatic Road Crack Techniques," 2023 2nd International Conference on Automation,
Detection and Characterization." IEEE Transactions on Computing and Renewable Systems (ICACRS), Pudukkottai, India,
Intelligent Transportation Systems 14.1 (2013): 155–168. 2023, pp.18261831,doi:10.1109/ICACRS58579.2023.10404670.

[8] Ren, Junhua, et al. "Automatic Pavement Crack Detection [17] R. Aruna, M. Prabu, S. Ananthi, V. C. Bharathi, R. Sathya and
B. Suchithra, "Vision based Cassava Plant Leaf Disease Classification
Fusing Attention Mechanism." Electronics 11.21 (2022): 3622.
using Machine Learning Techniques," 2023 2nd International
Conference on Automation, Computing and Renewable Systems
(ICACRS), Pudukkottai, India, 2023, pp. 959-964, doi:
[9] Sarhadi, A., Ravanshadnia, M., Monirabbasi, A., & 10.1109/ICACRS58579.2023.10404468.
Ghanbari, M. "Optimizing Concrete Crack Detection: An
Attention-Based SWIN U-Net Approach." IEEE Access, 2024.

[10] Shu, Jiangpeng, et al. "An Active Learning Method with

Difficulty Learning Mechanism for Crack Detection." Smart
Structures and Systems 29.1 (2022): 195-206.

[11] Vrochidou, Eleni, et al. "Towards Robotic Marble Resin

Application: Crack Detection on Marble Using Deep Learning."
Electronics 11.20 (2022): 3289.

[12] Varadharajan, S., S. Jose, K. Sharma, L. Wander, and C.

Mertz. "Vision for Road Inspection." In Proceedings of 2014
IEEE Winter Conference on Applications of Computer Vision,
2014, pp. 115–122.

[13] Yang, Yalong, et al. "Research on Pavement Crack

Detection Algorithm Based on Deep Residual Unet Neural
Network." Journal of Physics: Conference Series 2278.1 (2022).
IOP Publishing.

© 2025, IJSREM | www.ijsrem.com DOI: | Page 6