IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 4, December 2024, pp. 4480~4487

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i4.pp4480-4487  4480

Pothole detection model for road safety using computer vision

and machine learning

Vijaykumar S. Bidve1, Kiran S. Kakade2, Rahul H. Bhole3, Pakiriswamy Sarasu4, Ashfaq Shaikh5,
Pradnya Samit Mehta6, Santosh P. Borde7, Shailesh O. Kediya8
1
School of Computer Science and Information Technology, Symbiosis Skills and Professional University, Pune, India
2
Symbiosis Institute of Management Studies, Symbiosis International (Deemed University), Pune, India
3
MIT Art, Design and Technology University, Pune, India
4
Department of Computer Science and Engineering, Chennai Institute of Technology, Chennai, India
5
Department of Information Technology, M. H. Saboo Siddik College of Engineering, Mumbai, India
6
Department of Computer Science and Engineering-AI, Vishwakarma Institute of Information Technology, Pune, India
7
Student Progression and Industry Relations Office, JSPM’s Rajarshi Shahu College of Engineering, Pune, India
8
School of Logistics and Supply Chain Management, Symbiosis Skills and Professional University, Pune, India

Article Info ABSTRACT

Article history: Potholes pose significant threats to vehicular movement, causing damage to
vehicles and risking the safety of drivers and pedestrians. The escalating
Received Feb 24, 2024 issue of potholes has led to substantial financial losses for vehicle owners
Revised Jun 15, 2024 and drivers. Traditional methods of pothole detection are impractical,
Accepted Jun 21, 2024 necessitating an innovative approach. The study focuses on implementing a
detection system capable of accurately identifying potholes, empowering
vehicles to adapt their speed or halt to prevent damage. The transformative
Keywords: solution presented in this research leverages cutting-edge technologies,
specifically computer vision and machine learning, aiming to enhance road
Computer vision; safety and streamline maintenance efforts. By addressing the
Intelligent transportation interdependence of modern civilization on road networks, the Pothole
systems; Detection Model promises improved road safety, efficient maintenance
Machine learning; practices, and the emergence of an era in intelligent transportation systems.
Pothole; The integration of technology into transportation infrastructure highlights the
Vehicular safety proactive measures needed to combat road imperfections, ensuring a safer
and more efficient road network for the benefit of society.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Vijaykumar S. Bidve
School of Computer Science and Information Technology, Symbiosis Skills and Professional University
Kiwale, Pune, India
Email: [email protected]

1. INTRODUCTION
Potholes on roads are hard to pass for vehicles as they cause much damage to them. As each and
every pothole cannot be converted to a proper road, it is better to make a pothole detection model by which
vehicles are able to detect potholes so it can lower down its speed or stop itself [1], [2]. So, no harm will be
caused to the vehicle and it directly saves an enormous amount of money. Infrastructure adds significantly to
a country's economic overall growth [3], [4]. Road surface of cement and other types are commonly used as
mobility facilities in the world [5].
Potholes, irregular manholes and other irregularities are examples of roadways. Potholes can arise as
a result of poor construction, poor design that causes groundwater to gather [6]. Annually, potholes do a
significant damage to both people and their assets. Potholes immediately affect and disappoint significant
amount of people [7], [8]. Approximately drivers have spent over Rs. 10,000 crores in five to seven years on

Journal homepage: http://ijai.iaescore.com

Int J Artif Intell ISSN: 2252-8938  4481

automobiles to repair damage caused by potholes. Each motorist incurred a cost of almost Rs. 10,000 on an
average. The survey in India states that around 5,000 accidents on an average takes place. Governments also
spends a huge amount of money in repairing these potholes. So, vehicles need to get upgraded by pothole
detection model so it lowers it speed or stops [9], [10].
The road infrastructure is a critical component of urban and rural environments, serving as the
arteries that facilitate the movement of people and goods [11]. However, the inevitable wear and tear on road
surfaces, exacerbated by factors such as weather conditions, heavy traffic, and insufficient maintenance, give
rise to a pervasive and potentially hazardous issue like potholes [12]. Potholes not only contribute to a
degradation of the quality of roads but also pose significant risks to both vehicles and pedestrians. In
response to this challenge, the pothole detection project has been initiated to develop an innovative solution
leveraging computer vision and machine learning technologies [13], [14].
The intricate web of roads that crisscrosses urban and rural landscapes is the backbone of modern
civilization, facilitating the movement of people and goods [15], [16]. However, this intricate network is not
impervious to the relentless forces of nature and human activity. Potholes, those ubiquitous road
imperfections, emerge as an unintended consequence, threatening the integrity of our transportation
infrastructure and compromising the safety of road users. The advent of technology, particularly in the realms
of computer vision and machine learning, presents a unique opportunity to address this pervasive challenge.
The pothole detection project, at the intersection of innovation and infrastructure, embarks on a mission to
enhance road safety, streamline maintenance efforts, and herald a new era in intelligent transportation
systems [17]‒[19]. In current scenario, many vehicles and their drivers are facing issues regarding the
potholes on road which are causing much damage to cars and this leads to a loss which are covered by huge
amounts of money. To avoid this there are certain actions required which includes potholes detection, lower
the speed of vehicle or to stop, overcome the issues related to potholes, save one’s capital expenditure.

2. METHOD
2.1. Literature review
In this section state of art literature review is presented to know the current status developments in
the field of pothole detection. The main focus of the literature review is to know technologies used for
pothole detection, to find out the gaps and to work for the fulfillment of the gaps. The work tries to consider
the latest available developments in the field of pothole detection for the literature review.
Bučko et al. [20] discussed the omnipresence of road discrepancies like potholes and focused on the
pivotal role of early detection of potholes in prompt maintenance and accident prevention. The study uses
you only look once (YOLO)v3, a computer vision model, for automatic pothole detection. The impact of
adverse conditions including weather and light is also observed for detection accuracy of potholes. The
author presented two-fold structure including dataset creation and utilization of YOLOv3 and sparse
region-based convolutional neural networks (R-CNN). YOLOv3 is a competitive architecture, delivering
commendable results with lower hardware demand. YOLOv3 showcases superior performance in brighter
light conditions. Sparse R-CNN gives advantages in low light setting. This study works for comprehensive
dataset and establishes methodology for computer vision-based object detection under adverse condition.
Asad et al. [21] investigated asphalt pavement distress, particularly emphasizing pothole detection's
significance for roadway safety and daily commuting. Authors explored deep learning models deploying and
evaluating various frameworks on edge devices, showcasing Tiny-YOLOv4, and YOLOv5 models with
impressive mean average precision (mAP) scores. Authors explored that, despite high mAP, YOLOv5
exhibited limitations in distant pothole detection and misclassification, leading to the endorsement of
YOLOv4 for accuracy and Tiny-YOLOv4 for real-time detection. This approach promises swift
infrastructure repairs, potentially integrating global positioning system (GPS) for precise failure location
identification. Author future plan includes detecting various pavement distresses, road classifications, and
improving accuracy through real-time deployment modifications.
Hassan et al. [22] addressed the pivotal task of automating pothole detection in road maintenance. In
case of traditionally manual visual inspections, integrating machine learning models presents challenges in
producing generalizable models due to diverse image conditions. Using convolutional neural networks
(CNN) models like faster R-CNN and Inception V2, authors explored four datasets encompassing various
image sizes, distances, lighting, and environmental factors. Highlighting the impact of pothole distance and
potential false identifications model is extended to detect multiple pavement defects. The significance of this
study is to speed up and reduce the cost of pavement inspections by deploying the trained model for real-time
detection. Author’s future plan aims to leverage larger datasets, including commercial and drone-captured
images, necessitating robust pre-processing and increased computational resources for model refinement and
hyperparameter tuning.

Pothole detection model for road safety using computer vision and machine learning (Vijaykumar S. Bidve)
4482  ISSN: 2252-8938

Kumar et al. [23] highlighted the importance of detecting road imperfections, particularly potholes,
to enhance road maintenance and ensure vehicle safety. Authors noticed various image processing techniques
attempting automated monitoring, yet challenges persist due to diverse road conditions and pothole
variations. Addressing this, the study employed CNN to analyze digital images, achieving precise
assessments of pothole depth, area, and shape. This approach aims to enhance the accuracy and efficiency of
road condition detection, crucial for mitigating vehicle damage, fuel consumption, passenger inconvenience,
and reducing road accidents stemming from poor road conditions.
Zhang and Hamdulla [24] explored the importance of identifying road potholes, essential for
real-time vehicle control and road maintenance. Potholes, as unpredictable road features, significantly impact
driving conditions, vehicle damage, and road safety worldwide. With growing concern due to accidents and
advancements in machine learning and intelligent driving technologies, the demand for precise pothole
detection has increased. Authors investigated existing pothole detection methods, acknowledging the
evolving landscape of machine learning and intelligent driving, highlighting the necessity for robust pothole
detection strategies to safeguard lives and streamline travel.
Seetha [25] addressed the global issue of potholes on roads, emphasizing their potential for vehicle
damage and road safety risks. Leveraging deep learning algorithms, the authors presented a methodology to
employ faster R-CNN and YOLOv3 models for early pothole detection using images and videos, mitigating
accident probabilities. Authors considered the challenges like variable lighting and data noise, highlighting the
efficacy of deep learning in accurate pothole identification. The comparative analysis shows the effectiveness of
these algorithms in enhancing road maintenance, reducing vehicle damage, and augmenting safety. The proposed
model facilitates efficient road repairs and offers a user-friendly complaint system via an online platform,
indicating pothole locations from uploaded images. Author’s future work considers integration of GPS for
real-time detection and driver alerts, enhancing the system's usability and accuracy in detecting pavement defects.
Pramanik et al. [26] addressed the consequences of road accidents, attributing a significant cause to
potholes. The author’s main focus is on image processing techniques, specifically transfer learning, the
research aims to identify potholes using visual geometry group (VGG16) and ResNet50 neural network
models. Prioritizing road safety, the work aims to enable the Road and Highway Department (RHD) with a
computer vision-based solution for efficient pothole recognition. After employing various image
preprocessing methods, the proposed models give better accuracy. The findings claim that the models
developed for pothole identification and road safety enhancement are highly accurate.
Ma et al. [27] highlighted the prolonged use of computer vision in 3-D road imaging and pothole
detection. The authors identified gap in systematic surveys on cutting-edge techniques, especially deep
learning models. The author introduces 2-D/3-D road data acquisition systems and reviews state-of-the-art
computer vision algorithms for pothole detection. The current challenges are addressed in the article by
forecasting a transition from older methods to CNNs for comprehensive road assessment systems. This work
is important as a reference for advancing next-generation road condition assessment technologies.
Kandoi et al. [28] addressed the critical safety concerns in road transport due to poor maintenance,
particularly focusing on the extensive issue of potholes leading to accidents worldwide. The proposed
solution utilizes civilian’s mobile sensors and image-based methods, employing machine learning for
real-time pothole detection. System sends notification to authorities through a web-based portal to facilitate
prompt action, while the solution integrates features like location tagging, prioritization, and forecasting
potential pothole issues. It updates the time series data to enhance system's efficiency, offering a
comprehensive approach to address and prevent road accidents, emphasizing the importance of timely road
maintenance for global traveler safety.
Singh et al. [29] highlighted the detrimental impact of rough roads on travel quality, vehicle
maintenance, and safety. Authors focused on vision-based techniques for road condition detection. The authors
employed a transfer learning-based inception ResNetV2 CNN model for road images. The authors made
categories in the form of plane road, large pothole, and small pothole. The work evaluates model performance
through training and testing loss and compared with CNN and support vector machine models using the same
dataset. This work mainly contributes to road safety enhancement through vision-based classification methods.
Obreja and Dobrea [30] investigated the key interests and investment in cognitive systems and
artificial intelligence. The work focused on creating a framework for pothole identification in real-time. The
study captures and processes city images to identify asphalt potholes using video camera and Jetson Nano.
The envisioned outcome used to recognize hazardous potholes and communicating road malfunctions to a
central information hub to prevent traffic accidents. The authors mainly signify the convergence of AI
applications and edge computing to offer promising solution for proactive road safety measures.
The reviewed literature is summarized in the Table 1 to find out the gaps of existing literature. The
Table 1 presenting summary of literature survey contains the key attributes including title of the paper,

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4480-4487

Int J Artif Intell ISSN: 2252-8938  4483

author, technologies used, shortcomings. This Table 1 is used as a base to know the gaps and define the
problem statement to resolve the same.
The summary of the literature survey shows that the most of the paper reviewed are either working
with image processing or machine learning algorithms. In some of the papers other techniques are also used.
The accuracy of the proposed models is not up to the mark in most of the reviewed papers. The hardware
setup required by the system is also not explained in details by any of the authors. Hence this work
considered all the gaps of existing literature to develop a system and to get results with high precision.

Table 1. Summary of literature review

Sr. No. Title Author Methodology Limitation
1 Computer vision based pothole detection Bučko et al. [20] YOLOv3, R-CNN Main focus is on computer
under challenging conditions vision-based object detection
2 Pothole detection using deep learning: a real- Asad et al. [21] Tiny-YOLOv4, The edge device frameworks
time and ai-on-the-edge perspective YOLOv5 are the considered
3 Pothole detection under diverse conditions Hassan et al. [22] R-CNN, Inception V2 Lagging in model refinement
using object detection and hyperparameter tuning
4 A vision-based pothole detection using CNN Kumar et al. [23] Image processing, Accuracy and precision
mode CNN discussed
5 Research on pothole detection method for Zhang and Machine learning Result and implementation not
intelligent driving vehicle Hamdulla [24] discussed in depth
6 Intelligent deep learning-based pothole Seetha et al. [25] R-CNN, YOLOv3 Deep learning algorithm
detection and alerting system implementation is not
discussed in details
7 Detection of potholes using convolutional Pramanik et al. VGG16, ResNet50 Image processing methods are
neural network models: a transfer learning [26] not discussed in details
approach
8 Computer vision for road imaging and pothole Ma et al. [27] 3-D road imaging Deep learning models are not
detection: a state-of-the-art review of systems explained in details
and algorithms
9 Pothole detection using accelerometer and Kandoi et al. [28] Machine learning Machine Learning algorithm is
computer vision with automated complaint not explained in details
redressal
10 Classification of potholes using convolutional Singh et al. [29] ResNetV2 CNN Implementation details are not
neural network model: a transfer learning provided
approach using inception resnetV2
11 Developing a versatile framework needed to Obreja et al. [30] AI, Edge computing Implementation details are not
identify pothole in asphalt using deep learning provided
algorithms

2.2. Working
The road conditions, notably potholes, pose significant risks, causing accidents, vehicle damage, and
financial losses. Potholes stem from various factors: subpar materials, flawed design allowing water
accumulation, and weather conditions leading to cracks and ice formation. These gaps in road surfaces result
in severe damage to vehicles, often necessitating costly repairs and endangering lives. The existing repair
efforts fall short, unable to match the magnitude of the problem. The inadequate detection and rectification of
potholes exacerbate their impact on road safety and infrastructure. To address this issue, a pothole detection
model integrated into vehicles is proposed. This model aims to detect road potholes in real-time, triggering
vehicle responses such as speed reduction or automatic stops to prevent accidents and minimize vehicle
damage. The goal is to leverage technology and machine learning algorithms to proactively identify potholes,
ensuring the safety of drivers and passengers while curbing the financial burden caused by road damage. By
developing a pothole detection model capable of real-time identification and automatic response, this project
aims to significantly reduce vehicle damage, prevent accidents, and enhance road safety. This technological
advancement seeks to revolutionize road maintenance, mitigating the substantial economic and human costs
currently associated with pothole-related incidents.

2.2.1. Objectives
This research work aims to develop an advanced pothole detection system using a Raspberry Pi
remote-controlled (RC) model. The system is developed to identify road potholes autonomously. With the
help of machine learning algorithms, the efficiency and accuracy of pothole detection is enhanced, the
potential damage caused by road imperfectionsis reduced and overall road safety is improved. Key objectives
of the project are as given as:
− To develop an autonomous pothole detection system utilizing a Raspberry Pi-based RC car equipped with
a camera to identify and navigate around road potholes.

Pothole detection model for road safety using computer vision and machine learning (Vijaykumar S. Bidve)
4484  ISSN: 2252-8938

− To increase efficiency of pothole detection system using machine learning algorithms to mitigate damage
caused by road potholes and enhance road safety.
− Design a system that triggers automatic measures, such as slowing down or stopping the vehicle, upon
detecting potholes.
The main hardware components used in the implementation of the system are Raspberry Pi as a
main computing platform for processing and controlling the RC car, LM398 (motor driver) the component
enabling motor control for the RC car, power supply which provides necessary power for the Raspberry Pi
and RC car, RC car the vehicle for implementing the autonomous system, PI camera which captures real-time
images for pothole detection, USB cable used to connect components for data transfer, jumper wires and
male headers are the essential components for electrical connections. The software components used for the
system are integrated development environment (IDE) for code development, Jupyter notebook for
interactive environment for prototyping and testing, programming language-Python used for coding the
algorithms. OpenCV Library for image processing and computer vision tasks.
Using the hardware and software components the system is assembed and an algorithm is designed
to process camera images in real-time. During the data acquisition phase, the hardware model is established
and the cameras are strategically positioned in their functional locations. Utilizing OpenCV, the cameras
operate continuously, capturing a dataset comprising approximately 712 images of plain roads and roads
exhibiting potholes. These images are pivotal for training, testing, and refining the model to ensure its
accuracy and reliability. In the real-time live data context, the hardware setup generates essential data,
providing an efficient solution that allows drivers and users to relax without constant monitoring. The pothole
detection model actively observes the surroundings, facilitated by the continuous live video feed from the
cameras. Users have the flexibility to enable or disable the model as needed, granting them control over its
operation and ensuring seamless monitoring of road conditions.

2.2.2. Algorithm
Pothole detection relies on real-time video processing employing a CNN model. CNN is a machine
learning algorithm for the processing input images by assigning significant, learnable weights and biases to
diverse objects within the image, facilitating effective object differentiation. One notable advantage of CNN
lies in their reduced pre-processing demands compared to conventional classification algorithms. Traditional
methods require manually crafted filters, yet CNN possess the unique ability to autonomously learn these
filters and intricate features, significantly streamlining the pre-processing phase. Strength of this technology
is its adaptability and capacity to discern intricate patterns within visual data, making it a formidable tool for
pothole detection. By leveraging its inherent capability to learn features directly from the raw data, the CNN
model identifies and distinguishes potholes from their surrounding environment in real time. Its capacity to
automatically learn and adjust to various characteristics without explicit programming makes it a potent
solution for efficient and accurate pothole detection in live video streams. This attribute not only simplifies
the pre-processing steps but also enhances the model's capability to recognize and highlight potholes swiftly
and effectively within the continuous video feed.
The model initiates with input data, progressing through convolution and filter operations, including
loop-pattern, diagonal, and line filters, to generate a feature map. To prevent the loss of cells with high
features, padding is applied, ensuring a more precise function. Subsequently, the rectified linear unit (ReLU)
function introduces non-linearity, eliminating issues like the vanishing gradient problem. Pooling is then
employed, reducing image dimensionality while preserving essential features, enhancing accuracy with
reduced computational complexity. This iterative process repeats, refining outputs, culminating in an
appropriate output. Finally, an activation function enables the model to obtain the ultimate classification
output, utilizing these sequential steps to process and distil intricate visual data, leading to accurate and
efficient identification or classification within the input dataset.
The implemented functionalities are comprehensive, beginning with the hardware model capturing
live video data encompassing road conditions, particularly identifying potholes. This data retrieval initiates
the Pothole detection model, diligently examining the incoming road conditions to ascertain the presence of
potholes. Upon detecting a pothole, the system triggers mechanisms to reduce the vehicle's speed or halt its
movement altogether, ensuring safer navigation. Conversely, if no pothole is identified, the vehicle continues
its journey seamlessly. Leveraging CNN technology, this model guarantees the vehicle's operation on roads
without encountering hazardous potholes, boasting high precision and reliability in its assessments. Overall,
this amalgamation of hardware and detection models enables an intelligent response to road conditions,
ensuring vehicles navigate safely while actively averting potential pothole-related risks, underscoring the
efficacy and accuracy of the implemented system.

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4480-4487

Int J Artif Intell ISSN: 2252-8938  4485

3. RESULTS AND DISCUSSION

The implemented pothole detection system yielded commendable results, offering two significant
functionalities: retrieving live video data from hardware and accurately identifying potential potholes on
roads. The model's accuracy, a vital metric for autonomous vehicle operations on challenging terrains, proved
robust, high overall accuracy rate of 92%. This metric highlights the model's proficiency in differentiating
between road images with and without potholes. Moreover, precision and recall metrics underscored a
balanced performance. The precision rate of 91% indicates a low false-positive rate, ensuring accurate
identification of actual potholes, while the recall rate of 93% reflects the model's effectiveness in capturing a
high proportion of existing potholes.
Furthermore, the system's real-time processing capabilities were a notable feature, allowing for
instant pothole detection on video streams. This capability is critical for implementing proactive road
maintenance measures, enabling timely interventions to mitigate potential damage and hazards caused by
potholes. In essence, the developed pothole detection model, leveraging CNNs, demonstrated robustness and
accuracy in identifying and localizing potholes within various road scenarios. These promising results
indicate the potential of this system to contribute significantly to road safety, vehicle maintenance, and
proactive infrastructure management.

4. CONCLUSION
The proposed system of pothole detection with live video capture from the hardware model, using
CNN proved to be a reliable and precise method. This work offeres effective observation of potholes,
presenting a significant advancement in perceiving road anomalies in real-time through live video feeds. The
implementation of CNN for pothole detection signifies a significant stride, particularly in the realm of
self-driving cars, especially in challenging conditions. This research contributes significantly to safer and
more resilient urban infrastructure, paving the way for improved road safety and enhanced driving
experiences. The future scope involves augmenting its capabilities by integrating additional models for
pedestrian and obstacle detection, enhancing overall environmental visualization.

REFERENCES
[1] C. Katrakazas, M. Quddus, W.-H. Chen, and L. Deka, “Real-time motion planning methods for autonomous on-road driving:
State-of-the-art and future research directions,” Transportation Research Part C: Emerging Technologies, vol. 60, pp. 416–442,
Nov. 2015, doi: 10.1016/j.trc.2015.09.011.
[2] S. A. Bagloee, M. Tavana, M. Asadi, and T. Oliver, “Autonomous vehicles: challenges, opportunities, and future implications for
transportation policies,” Journal of Modern Transportation, vol. 24, no. 4, pp. 284–303, Dec. 2016, doi: 10.1007/s40534-016-0117-3.
[3] S. Sattar, S. Li, and M. Chapman, “Road surface monitoring using smartphone sensors: A Review,” Sensors, vol. 18, no. 11, Nov.
2018, doi: 10.3390/s18113845.
[4] Y. Jo and S. Ryu, “Pothole detection system using a black-box camera,” Sensors, vol. 15, no. 11, pp. 29316–29331, Nov. 2015,
doi: 10.3390/s151129316.
[5] A. A. Alhussan, D. S. Khafaga, E.-S. M. El-Kenawy, A. Ibrahim, M. M. Eid, and A. A. Abdelhamid, “Pothole and plain road
classification using adaptive mutation dipper throated optimization and transfer learning for self driving cars,” IEEE Access, vol.
10, pp. 84188–84211, 2022, doi: 10.1109/ACCESS.2022.3196660.
[6] M. P. B and S. K.C, “Enhanced pothole detection system using YOLOX algorithm,” Autonomous Intelligent Systems, vol. 2, no.
1, Aug. 2022, doi: 10.1007/s43684-022-00037-z.
[7] D. J., S. D. V., A. S A., K. R., and L. Parameswaran, “Deep learning based detection of potholes in Indian roads using YOLO,” in
2020 International Conference on Inventive Computation Technologies (ICICT), Feb. 2020, pp. 381–385, doi:
10.1109/ICICT48043.2020.9112424.
[8] P. A. Chitale, K. Y. Kekre, H. R. Shenai, R. Karani, and J. P. Gala, “Pothole detection and dimension estimation system using
deep learning (YOLO) and image processing,” in 2020 35th International Conference on Image and Vision Computing New
Zealand (IVCNZ), Nov. 2020, pp. 1–6, doi: 10.1109/IVCNZ51579.2020.9290547.
[9] Q. Li, Y. Shi, Q. Liu, and G. Liu, “Deep learning-based pothole detection for intelligent transportation: A YOLOv5 approach,”
International Journal of Advanced Computer Science and Applications, vol. 14, no. 12, 2023, doi: 10.14569/IJACSA.2023.0141242.
[10] K. R. Ahmed, “Smart pothole detection using deep learning based on dilated convolution,” Sensors, vol. 21, no. 24, Dec. 2021,
doi: 10.3390/s21248406.
[11] S.-Y. Lee, T. H. M. Le, and Y.-M. Kim, “Prediction and detection of potholes in urban roads: Machine learning and deep learning
based image segmentation approaches,” Developments in the Built Environment, vol. 13, Mar. 2023, doi: 10.1016/j.dibe.2022.100109.
[12] F. Ozoglu and T. Gökgöz, “Detection of road potholes by applying convolutional neural network method based on road vibration
data,” Sensors, vol. 23, no. 22, Nov. 2023, doi: 10.3390/s23229023.
[13] E. N. Ukhwah, E. M. Yuniarno, and Y. K. Suprapto, “Asphalt pavement pothole detection using deep learning method based on
YOLO neural network,” in 2019 International Seminar on Intelligent Technology and Its Applications (ISITIA), Aug. 2019, pp.
35–40, doi: 10.1109/ISITIA.2019.8937176.
[14] S.-S. Park, V.-T. Tran, and D.-E. Lee, “Application of various YOLO models for computer vision-based real-time pothole
detection,” Applied Sciences, vol. 11, no. 23, Nov. 2021, doi: 10.3390/app112311229.
[15] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: unified, real-time object detection,” in 2016 IEEE
Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2016, pp. 779–788, doi: 10.1109/CVPR.2016.91.
[16] H. Song, K. Baek, and Y. Byun, “Pothole detection using machine learning,” in Advanced Science and Technology, Feb. 2018,
pp. 151–155, doi: 10.14257/astl.2018.150.35.

Pothole detection model for road safety using computer vision and machine learning (Vijaykumar S. Bidve)
4486  ISSN: 2252-8938

[17] M. K. Tripathi, D. Baswaraj, S. Deshmukh, K. Misal, N. P. Bhosle, and S. M. Sangve, “Pothole detection in bituminous road
using convolutional neural network with transfer learning,” Indonesian Journal of Electrical Engineering and Computer Science,
vol. 33, no. 3, pp. 1950–1957, Mar. 2024, doi: 10.11591/ijeecs.v33.i3.pp1950-1957.
[18] C. Senigalakuruba and S. Pabba, “Pothole recognition using convolution neural networks and transfer learning,” IAES International
Journal of Artificial Intelligence (IJ-AI), vol. 12, no. 3, pp. 1204–1209, Sep. 2023, doi: 10.11591/ijai.v12.i3.pp1204-1209.
[19] G. K. and T. S., “Novel deep learning model for vehicle and pothole detection,” Indonesian Journal of Electrical Engineering and
Computer Science, vol. 23, no. 3, pp. 1576–1582, Sep. 2021, doi: 10.11591/ijeecs.v23.i3.pp1576-1582.
[20] B. Bučko, E. Lieskovská, K. Zábovská, and M. Zábovský, “Computer vision based pothole detection under challenging
conditions,” Sensors, vol. 22, no. 22, Nov. 2022, doi: 10.3390/s22228878.
[21] M. H. Asad, S. Khaliq, M. H. Yousaf, M. O. Ullah, and A. Ahmad, “Pothole detection using deep learning: a real-time and AI-on-
the-edge perspective,” Advances in Civil Engineering, vol. 2022, pp. 1–13, Apr. 2022, doi: 10.1155/2022/9221211.
[22] S. Hassan, D. O’Sullivan, and S. Mckeever, “Pothole detection under diverse conditions using object detection models,” in The
International Conference on Image Processing and Vision Engineering, 2021, pp. 128–136, doi: 10.5220/0010463701280136.
[23] P. Kumar, Pooja, N. Chauhan, and N. Chaurasia, “A vision-based pothole detection using CNN model,” SN Computer Science,
vol. 4, no. 6, Sep. 2023, doi: 10.1007/s42979-023-02153-w.
[24] F. Zhang and A. Hamdulla, “Research on pothole detection method for intelligent driving vehicle,” in 2022 3rd International
Conference on Pattern Recognition and Machine Learning (PRML), 2022, pp. 124–130, doi: 10.1109/PRML56267.2022.9882211.
[25] D. M. Seetha, “Intelligent deep learning-based pothole detection and alerting system,” International Journal of Computational
Intelligence Research (IJCIR), vol. 19, no. 1, pp. 25–35, Jun. 2023, doi: 10.37622/IJCIR/19.1.2023.25-35.
[26] A. Pramanik, M. H. I. Bijoy, and M. S. Rahman, “Detection of potholes using convolutional neural network models: a transfer
learning approach,” in 2021 IEEE International Conference on Robotics, Automation, Artificial-Intelligence and Internet-of-
Things (RAAICON), Dec. 2021, pp. 73–78, doi: 10.1109/RAAICON54709.2021.9929623.
[27] N. Ma et al., “Computer vision for road imaging and pothole detection: a state-of-the-art review of systems and algorithms,”
Transportation Safety and Environment, vol. 4, no. 4, 2022, doi: 10.1093/tse/tdac026.
[28] A. Kandoi, H. Agarwal, A. Revadekar, and B. N. Chaudhari, “Pothole detection using accelerometer and computer vision with
automated complaint redressal,” in 2021 3rd International Conference on Signal Processing and Communication (ICPSC), May
2021, pp. 257–264, doi: 10.1109/ICSPC51351.2021.9451762.
[29] S. Singh, R. Chhabra, and A. Moudgil, “Classification of potholes using convolutional neural network model: a transfer learning
approach using inception resnetV2,” in 2023 2nd Edition of IEEE Delhi Section Flagship Conference (DELCON), Feb. 2023, pp.
1–5, doi: 10.1109/DELCON57910.2023.10127302.
[30] M.-E. Obreja and D.-M. Dobrea, “Developing a versatile framework needed to identify pothole in asphalt using deep learning
algorithms,” in 2022 E-Health and Bioengineering Conference (EHB), Nov. 2022, pp. 1–4, doi: 10.1109/EHB55594.2022.9991727.

BIOGRAPHIES OF AUTHORS

Dr. Vijaykumar S. Bidve is Associate Professor at School of CSIT, Symbiosis

Skills and Professional University, Kiwale, Pune, Maharashtra, India. He Holds a Ph.D.
degree in Computer Science & Engineering with specialization in Software Engineering. His
research areas are software engineering, machine learning, cyber security. He has published
10 patents and more than 45 research articles in national and international journals. He is a life
member of ISTE. He is working as an expert for various subjects. Also, he has worked as a
reviewer for various conferences and journals. He can be contacted at email:
[email protected].

Dr. Kiran S. Kakade is Associate Professor at Symbiosis Institute of

Management Studies, Symbiosis International (Deemed University), Pune, India. He has
completed his graduation in Management Science and post-graduation in MBA (HR), LL.M
(Law), and Master in Computer Application (MCA) from the prestigious University of Pune.
His research areas include artificial intelligence in HR and law. He has published 11 patents
and over 50 research papers published in esteemed ABDC and Scopus publications. He has
significant contributions to academia and industry, notably at esteemed institutions like TISS,
TIMSR, GNIMS, and MIT World Peace University. He can be contacted at email:
[email protected].

Prof. Rahul H. Bhole is an Assistant Professor at Department of Information

Technology, School of Computing, MIT-ADT University Pune, whose expertise lies at the
intersection of cutting-edge technologies and safeguarding sensitive data. As a dedicated
researcher in cloud computing, federated learning, and data security, his work illuminates the
evolving landscape of digital privacy. With a passion for unravelling complexities, he
navigates the realms of cloud infrastructures, exploring innovative approaches to optimize
efficiency without compromising security. His commitment to advancing federated learning
techniques showcases her deep understanding of collaborative machine learning paradigms.
He can be contacted at email: [email protected].

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4480-4487

Int J Artif Intell ISSN: 2252-8938  4487

Dr. Pakiriswamy Sarasu is Professor at Department of Computer Science and

Engineering, Chennai Institute of Technology, Chennai, India. She holds a Ph.D. Degree in
Computer Science and Engineering. She did her masters in Embedded Systems Technology
and Bachelors in Computer Science engineering. Her research areas include chaotic systems,
cryptography, and autonomous vehicle. She played a major role in creating innovation
entrepreneurial ecosystem and more than 120 startups are created under her guidance and
mentorship. One patent is granted for her as one of the inventors. She can be contacted at
email: [email protected].

Dr. Ashfaq Shaikh is working as Assistant Professor at M. H. Saboo Siddik

College of Engineering, Byculla, Mumbai, India. He holds Ph.D. in Computer Engineering
with a specialization in Big Data Analytics, Machine Learning, Recommendation System,
Information and Cyber security. His passion for teaching and innovation contribution resulted
in winning several awards and recognition such as mastek deep blue winner in 2017, AICTE
Best Team Award in Smart India Hackathon in 2018, and best faculty award in year 2021. He
can be contacted at email: [email protected].

Prof. Pradnya Samit Mehta is working as Assistant Professor at the

Vishwakarma Institute of Information Technology Pune. She has more than 15 years of
teaching experience in Computer Engineering and 2 years of Industrial Experience. Her
current research interests include artificial intelligence, deep learning, high performance
computing, and machine learning. She has published more than 20 papers in national and
international journals and conferences, 7 book chapters, and 7 patents, 2 are granted. She has
also published 3 books on Python Libraries. She can be contacted at email:
[email protected].

Dr. Santosh P. Borde is working as Assistant Director for Student Progression

and Industry Relations Office at JSPM’s Rajarshi Shahu College of Engineering, Pune, India.
He has 21 years of experience in the field of Education. He handles various responsibilities
towards training and development, corporate relations and alumni relations. His area of
interest is in the field of human computer interaction. He worked on e-learning model using
human computer aspects of usability during his Ph.D. work. He can be contacted at email:
[email protected].

Dr. Shailesh O. Kediya is Associate Professor at Symbiosis Skills and

Professional University, Kiwale, Pune, Maharashtra, India. He Holds a Ph.D. degree in
Logistics Management. His research areas are innovative and disruptive technologies,
business management. He has published 2 international patents, awarded with 12 copyright
and written 8 books, and more than 50 research articles in national and international journals.
He is a member of IEEE, SEBI, and NISM. Also, he has been editor of four international
journals and worked as a reviewer for various conferences and journals. He can be contacted
at email: [email protected].

Pothole detection model for road safety using computer vision and machine learning (Vijaykumar S. Bidve)