DEPARTMENT OF CIVIL ENGINEERING, TKMCE

Automated Pavement Distress Detection in Road Maintenance

Management: Necessity, Innovations, and Challenges - A Literature
Review
Shernaz V Seyad
M-Tech Transportation Engineering, Department of Civil Engineering, TKM College of Engineering, Kollam,
Kerala.

Abstract: This literature review addresses the pressing need for automated pavement distress detection,
with a specific focus on potholes, in the context of road maintenance management systems. Current
methods heavily rely on manual inspection, incurring time and financial costs. The urgency for
automation is emphasized, considering the pivotal role of well-maintained asphalt pavements in
facilitating mobility and contributing to economic growth. Existing research efforts, categorized into 3D
reconstruction-based, vibration-based, and vision-based methods, are explored, highlighting their merits
and drawbacks. The review outlines various methodologies for pothole detection, including an
automated approach employing a network of passenger vehicles, a hierarchical top-down method, the
identification of rut and pothole using a Multirotor Unmanned Aerial Vehicle (UAV), automated
assessment of potholes, cracks, and patches from road surface video clips using a two-stage approach,
and a real-time detection system utilizing thermal images and Convolutional Neural Networks (CNN).
Despite notable advancements, challenges persist, warranting continued refinement and innovation in
automated distress detection for efficient road maintenance practices.

Key words: Automated pavement distress detection, pothole detection, road maintenance, 3D
reconstruction, vibration-based methods, vision-based methods, challenges, innovation, road safety.

1. Introduction

The significance of well-maintained asphalt pavements cannot be overstated, considering their role in
facilitating mobility, fostering economic growth, and contributing to a country's overall development.
The correlation between a nation's economic prosperity and the quality of its road network underscores
the importance of maintaining well-functioning road surfaces. Neglected or poorly maintained roads can
impede mobility, elevate vehicle operating costs, escalate accident rates, and contribute to societal issues
such as poverty, poor health, and illiteracy in rural communities [5]. Therefore, consistent monitoring
and timely maintenance of asphalt pavement surfaces are imperative.

Asphalt pavements, however, are susceptible to various distresses, structural disorders that diminish
utility and serviceability. These distresses manifest as visible defects on the pavement surface,
encompassing block cracking, alligator cracking, transverse cracking, depressions, joint reflection
cracks, slippage, potholes, rutting, and shoving [5]. Among these, potholes emerge as a critical concern,
defined as depressions, holes or pits in the pavement surface resulting from the damage to the surfacing

1
material. In Figure. 1 we can see a few examples of potholes as variable sized bowl-shaped dumps in the
pavement surface.

Figure. 1: Different appearances of potholes in the pavement surface.

In the context of road maintenance management systems, the assessment of road surface distresses
stands as a pivotal task for formulating effective repair and maintenance strategies. Potholes, cracks, and
patches emerge as notable road surface distresses, demanding comprehensive evaluation, particularly in
regions like India (MORTH, 2004) [2]. The conventional approach involves distress data collection and
subsequent processing, employing various road distress data evaluation systems. These systems fall into
three categories based on the techniques employed: manual, sensor-based, and imaging-based [2].
Despite the increasing automation of distress data collection using imaging systems, the analysis of raw
video clips for distress assessment predominantly remains a manual process, incurring expenses, time,
and impeding road maintenance management efficiency.

In developed countries, extensive road networks necessitate robust maintenance programs to monitor
performance, predict future conditions, support investment planning, and identify rehabilitation needs.
The Long-Term Pavement Performance program(LTPP) in the United States exemplifies such
initiatives, emphasizing the importance of reliable measurements on distresses like cracks, potholes, and
rutting [3]. Due to the time-consuming, costly, and subjective nature of manually reviewing pavement
image and video data for defect detection, there is a growing emphasis on automating this process, with
a specific focus on potholes. Existing research in pothole detection can be categorized into 3D pavement
reconstruction methods employing technologies like laser scanning and stereo-vision, as well as
vibration-based approaches using acceleration sensors. Each approach, however, presents limitations in
terms of equipment costs, computational efforts, accuracy, and reliability [3].

Potholes, posing a significant threat to road safety and vehicle integrity, have attracted increased
attention due to their association with road accidents and fatalities. The inefficiency in detecting and
addressing potholes in a timely manner contributes to their persistence and the associated risks. While
traditional manual methods of pothole detection, known for being labor-intensive and time-consuming,
have prompted exploration of various techniques, including vibration-based methods, 3D reconstruction,
and vision-based methods, each technique exhibits certain limitations. This necessitates continuous
exploration of innovative approaches, emphasizing the need for ongoing improvement in pothole
detection methods. This literature review aims to delve into the diverse methods employed for pothole

2
detection and analysis, assessing their strengths, limitations, and potential for enhancing road
maintenance practices.

2. Urgency in Pavement Distress Detection: Necessity for Automation and Advancements in

Current Methods

Pavement distresses, visible defects on road surfaces, serve as symptomatic indicators of deterioration,
encompassing issues like cracks, potholes, and rutting. The inspection and maintenance of pavement
surfaces play a pivotal role in ensuring road safety and longevity. Road maintenance, particularly in
countries like India, is typically managed by government entities such as the Public Works Department
(PWD). However, the traditional methods of assigning dedicated personnel for surveying and
monitoring road conditions have proven to be tedious, time-consuming and financially burdensome.
This has led to a growing exploration of automation, especially in the domain of pavement distress
detection.

Traditionally, technicians review captured video data on computer screens, guided by personal
experiences and a distress manual categorizing severity and extent. This manual approach is further
compounded when digital inspection vehicles are unavailable, leading to surveys conducted manually
from vans at speeds of 8 to 15 mph with data collection assistance software [3].

Recent advancements include semi-automated crack detection using commercial software with digital
image processing algorithms, executed off-line and overnight. However, a notable limitation lies in the
incapacity of existing commercial software to automate the detection of various distress types such as
potholes, patching, and discoloration [3].

To assess pavement condition comprehensively, indices for distresses and roughness are essential, with
a surface distress index combined with a roughness index for an overall pavement condition rating.
While pavement roughness is automatically determined in real-time based on accelerometer and laser
scanner data, distress identification lacks similar efficiency [3].

Existing automated systems, like Pavemetrics, have made notable progress in detecting potholes using
3D data. Despite this, they lack sufficient accreditation for their performance [5]. Various methodologies
have been proposed for pothole detection, ranging from image processing algorithms to machine
learning and unsupervised techniques. However, these approaches encounter challenges in accurately
assessing pothole width, depth, and severity. Factors like dirt-filled potholes, shadows, and diverse
pavement conditions further complicate the detection process [5].

Potholes, a significant distress type, are particularly challenging for manual assessment due to their
bowl-shaped depressions. The identification process is time-consuming, tedious, and subjective,
influenced by raters' experience, leading to repeatability issues in the final assessment [3]. Recent
research has explored innovative solutions to enhance pothole detection accuracy. Spatio-temporal
saliency, the integration of ground-penetrating radar (GPR) data, and wavelet energy field methods have
shown promise [5]. However, challenges persist, such as detecting flat or bright potholes and addressing
the impact of blur and lighting conditions.

3
Technological advancements, including mobile applications for reporting road conditions, have been
introduced in countries like Malaysia [4], but they often lack the capability to assess specific geometric
details of distresses like ruts and potholes. These gaps highlight the need for new methods capable of
providing comprehensive information on distress geometry, including shape, perimeter, area, and depth,
to facilitate efficient road maintenance.

These challenges underscore the urgent need for automation in pavement distress detection. This
literature review specifically focuses on the automation of pothole detection and emphasizes the
effectiveness of automated detection compared to manual inspection. The goal is to contribute to the
evolution of more efficient and reliable pavement distress detection methodologies.

3. Current Research Efforts in Pothole Detection: Method Types and Drawbacks

Pothole detection is a critical aspect of pavement maintenance, and current research endeavors focus on
three main methodologies: 3D reconstruction-based, vibration-based, and vision-based methods [3], [2].
Each approach has its merits and drawbacks.

3.1. 3D Reconstruction-Based Methods:

Researchers have explored utilizing 3D point clouds generated by laser scanners or stereo-vision
algorithms for pavement assessment. Time-of-flight laser scanners and hybrid systems that incorporate
digital cameras have been employed for real-time 3D scanning to detect pavement distortions. However,
stereo-vision-based approaches face challenges in matching points due to irregular pavement textures,
making real-time implementation computationally demanding. Additionally, while laser scanning
systems offer real-time data collection, they still incur high equipment costs, limiting their practicality
[3]. The limitations of 3D-image based approaches, primarily associated with high equipment and
computation costs, emphasize the immediate need for more efficient and cost-effective solutions [2].

3.2. Vibration-Based Methods:

Vibration-based approaches utilize accelerometers to detect potholes by analyzing the vehicle's

mechanical responses. This method is cost-effective, requires small storage, and is suitable for real-time
processing. Projects like BusNet and Pothole Patrol integrate vibration sensors with GPS for mobile
road condition sensing [3]. However, these methods may yield incorrect results, such as misidentifying
bridge expansion joints as potholes. Potholes located in the center of a lane, untouched by vehicle
wheels, pose challenges for accurate detection but are crucial to capture for comprehensive pavement
defect assessment [3]. Vibration-based methods face criticism for compromised accuracy and reliability,
thereby limiting their applications in the comprehensive assessment of road distress [2].

3.3. Vision-Based Methods:

An alternative approach relies on 2D vision, using image or video data to identify potholes distinct from
the pavement background. While Karuppuswamy et al. [3] proposed a vision and motion system for
pothole detection and avoidance, this method's applicability is limited to simulated potholes larger than
2 feet in diameter and white in color, deviating from realistic pavement conditions. Further research is

4
needed to develop vision-based approaches that address the complexity and diversity of actual road
surfaces. 2D-video image-based methods, while proficient in singularly focusing on cracks or potholes,
fall short in addressing the practical need for the simultaneous detection and assessment of various
distress types [2].

In summary, while each method shows promise in automating pothole detection, challenges such as
computational demands, equipment costs, and the need for accurate real-time assessment highlight the
ongoing research efforts aimed at refining these methodologies for practical implementation in diverse
pavement environments. The literature points to a critical gap in the absence of methods capable of
efficiently automating the detection and measurement of potholes, cracks, and patches collectively from
real-life video clips.

In navigating the challenges within the realm of automated distress detection from real-life video clips,
especially those collected without artificial lighting systems, a delineation of obstacles becomes
apparent. The presence of extraneous elements, such as objects, road markings, and diverse shadows,
introduces noise in the data, posing a significant obstacle to the development of an effective automated
system [2]. Overcoming these challenges is imperative for advancing the accuracy and reliability of
distress assessments, streamlining the process to be more efficient and resource-effective. In conclusion,
addressing the current shortcomings and challenges is pivotal, and the development of a comprehensive
automated method capable of detecting and measuring various distress types from real-life video clips
holds the key to the advancement of road condition assessment practices.

4. Exploring Advances in Automated Methods: A Comprehensive Literature Review

Koch and Brilakis (2011) addresses automated pavement distress detection, focusing on potholes. The
methodology involves image segmentation, morphological thinning, elliptic regression, and texture
comparison. Figure 2(a) illustrates the procedural steps within the pothole detection model. The
Automated Pothole Detection Methodology sets out with the primary objective of evaluating the
feasibility of implementing a comprehensive framework for detecting, recognizing, spatially locating,
and evaluating pavement defects. This is achieved by harnessing a network of passenger vehicles
equipped with technology to simultaneously assess road conditions and their impact on drivers. Figure.2
(b) shows an image of a robot vehicle used for visual pavement sensing. The robot simulates an ordinary
passenger vehicle equipped with a high-speed camera mounted on the rear bumper.

The methodology involves a two-stage video processing approach, where high-speed algorithms in real-
time detect frames with potential pavement defects, followed by defect recognition algorithms analyzing
selected frames and tagging regions with identified defect types. A pattern recognition model is
employed, considering visual characteristics such as darker shadows, elliptical shape, and coarser
texture in potholes. The methodology incorporates image segmentation, shape extraction, and texture
comparison components. The methodology demonstrated an overall accuracy of 86% when
implemented their approach in a MATLAB prototype, tested it on a dataset of 120 pavement images,
and achieved promising results with an overall accuracy of 86%., with 82% precision and 86% recall
[3].

5
Figure.2: (a) Pothole detection model. (b) Robot vehicle for visual pavement sensing. The robot
simulates an ordinary passenger vehicle equipped with a high-speed camera mounted on the rear
bumper.

Among its advantages are the utilization of existing infrastructure, real-time processing for timely defect
identification, and a vision-based approach enhancing accuracy. However, challenges include reliance
on normal lighting conditions and manual training, suggesting potential enhancements through machine
learning and parameter optimization. Future directions involve exploring additional visual
characteristics, conducting parameter optimization, employing machine learning techniques,
implementing sensitivity analysis for robustness, and focusing on automated pothole detection in
pavement videos, considering depth and area estimations for assessment. The paper's strengths lie in its
systematic methodology, clear presentation of results, and acknowledgment of future directions. The
authors effectively address a real-world problem, recognize limitations, and propose avenues for
improvement. The comprehensive approach using image segmentation, shape extraction, and texture
comparison components contributes valuable insights to automated pavement defect detection, while
transparency about limitations enhances credibility.

Yousaf et al. (2018) presents a comprehensive and innovative approach to pothole detection and
localization using computer vision. The proposed approach presents a novel and hierarchical top-down
methodology for the detection and localization of potholes on asphalt pavement surfaces. The method
begins with a "Classification of Pavement Images" stage, employing a Bag of Words (BoW) approach
and supervised learning with Support Vector Machine (SVM) classification based on low-level features,
specifically scale-invariant feature transform (SIFT) features capturing significant points in pavement
images. This is followed by a "Pothole Localization" stage, utilizing a graph-based segmentation method
with normalized cuts to accurately identify and localize potholes based on homogeneity criteria. As
shown in Table.1, the approach demonstrates robustness to arbitrary shapes, illumination changes, and
textural variations in pavement images, achieving high accuracy (95.7%), precision (97%), and recall
(94.1%) in image classification. Despite potential computational overhead due to the SIFT feature
extraction, the methodology proves effective, outperforming existing methods in terms of accuracy,

6
precision, and processing time. The subjective and objective evaluations further confirm the
methodology's success, with 100% accurate localization in 72.3% of images and an introduced
localization index (ILoc) averaging 0.914 [5].
Table 1: Performance of pavement image categorization using BoW.

The proposed methodology's strengths lie in its independence from assumptions about pothole shapes,
illumination restrictions, and textural variations, rendering it a robust solution for categorizing asphalt
pavement images. The utilization of local features, specifically SIFT with a BoW, contributes to
superior performance. The paper offers both subjective and objective assessments, showcasing
remarkable outcomes and introducing an innovative objective evaluation criterion for pothole
localization, thereby augmenting the credibility of the proposed scheme. The study concludes by
suggesting potential applications for highway authorities in road surface analysis and maintenance,
including real-time implementation and incorporating geo-tagging features. Future directions involve
exploring more robust distress detection features, texture-based techniques for asphalt pavement
analysis, and extending the methodology to real-time applications and new dataset recordings. Overall,
the paper provides a valuable contribution to the field with its innovative methodology and thorough
evaluation, setting a foundation for enhanced road maintenance practices.
Saad and Tahar (2019) explores the application of UAV technology for assessing road conditions,
specifically rut and pothole detection. The presented methodology encompasses four key phases: site
reconnaissance and planning, data acquisition, data processing, and result and data analysis. The primary
objective is to assess road conditions, focusing on parameters like rut length, pothole size, and other
distress factors. The methodology employs UAV imagery for road pavement detection, allowing for a
comparison between actual measurements and those obtained from UAV images. The site
reconnaissance and planning phase involve selecting study areas, equipment (UAV quadcopter, digital
camera), and flight path planning. In data acquisition, the UAV captures images of road surfaces,
emphasizing areas with ruts and potholes, employing manual navigation and 70% image overlap. The
subsequent data processing phase utilizes Agisoft Photoscan for tasks such as photo alignment, cloud
building, and digital evaluation model (DEM) creation [4].

The advantages of this method include efficient data collection, detailed measurement accuracy, and
digital documentation for potential re-measurement. Drawbacks involve potential errors in UAV
measurements and manual navigation challenges. The materials and methods encompass various tools,
including thread, tape, and a UAV quadcopter with specific software. The conclusion highlights the
suitability of UAV imagery for pavement detection, offering a cost-effective and flexible solution.
Future work could explore LiDAR UAV technology to enhance data collection. This will allow for more
efficient road condition assessment in fully automatic mode. The study focuses on evaluating the effect
of different altitudes on resolution in rut and pothole extraction, utilizing structure-from-motion-based

7
photogrammetric software for data processing. The accuracy assessment is based on a comparison
between actual and measured data, concentrating on rut and pothole samples.

The presented methodology demonstrates the capability of multirotor UAV imagery for rut and pothole
extraction, providing detailed measurements of road conditions. The paper effectively communicates the
systematic approach, clearly detailing each phase of the methodology. The inclusion of an accuracy
assessment based on actual and measured data enhances the credibility of the results. However, the
paper lacks specific details on the UAV and camera specifications, flight parameters, and the nature of
the road conditions assessed, which could impact the reproducibility and generalizability of the findings.
Additionally, the conclusion emphasizes the suitability of UAV imagery for pavement detection but
does not thoroughly discuss potential limitations and challenges, such as weather conditions affecting
UAV flights or the need for skilled operators. Future work suggestions involving LiDAR UAV
technology and advancements in sensor technologies add valuable directions for further research.
Overall, the paper contributes to the growing field of UAV applications in road maintenance but would
benefit from more comprehensive technical details and a thorough discussion of limitations.

Huidrom et al. (2013) addresses the need for automated distress assessment in road infrastructure,
focusing on the detection and measurement of potholes, cracks, and patches in Indian highways. Using
various image processing techniques and heuristically derived decision logic based on distinctive visual
properties such as texture, shape, and dimension, the proposed method aims to enhance the efficiency
and accuracy of distress identification in road infrastructure [2].

The proposed method unfolds in two crucial stages. Initially, a swift Video Distress Frames Selection
(DFS) algorithm efficiently processes the video clip database, automating the separation of frames with
distress from those without. Subsequently, the Critical Distress Detection, Measurement, and
Classification (CDDMC) algorithm come into play, facilitating the automated detection and assessment
of potholes, cracks, and patches in a single pass. Comprising five steps—Image Enhancement, Image
Segmentation, Visual Properties Extraction, Detection and Classification by Decision Logic, and
Quantification—the CDDMC algorithm relies on three key visual properties: standard deviation (STD),
circularity (CIRC), and average width (W). The method categorizes frames into those with potholes,
cracks, patches, and frames without critical distress, presenting the information in a printable format. In
Figure 3, image in column (a) is original images with potholes, cracks and/or shadows; column (b) is the
binary processed images obtained after applying the CDDMC algorithm in column (a) images and
column (c) is the extracted visual properties of critical objects in column (b) images, where CO1 = First
biggest critical object (CO), CO2 = Second biggest CO, CO3 = Third biggest CO [2].

8
 Figure 3: (a) Original image with potholes; (b) processed image; (c) extracted visual properties [2]

The methodology showcases notable advantages, including a substantial speedup, approximately 60

times faster than manual methods, coupled with an impressive overall accuracy of 97% for potholes,
94% for cracks, and 90% for patches. The resource savings in terms of time, money, and manpower are
significant. However, certain drawbacks exist, such as the focus on a limited set of distress types,
potential challenges in distinguishing distresses from visually similar features, and the need for future
enhancements to encompass additional distress types and refine the classification process.

The methodology delivers notable advantages, offering a substantial speedup over manual methods with
impressive overall accuracy rates. The inclusion of visual properties like standard deviation, circularity,
and average width enhances the efficiency and accuracy of distress identification. Despite limitations,
such as a focus on a limited distress set and challenges in distinguishing visually similar features, the
paper recognizes the need for future enhancements. The method showcases efficiency and accuracy in
automated distress assessment, leading to significant time and resource savings. The incorporation of
distinctive visual properties bolsters detection reliability, and future directions emphasize algorithm
development for enhanced classification and severity measurement. Overall, the paper contributes to
road maintenance management by presenting a robust method for automated distress assessment in real-
life video clips, specifically tailored for Indian highways.

Aparna et al. (2022) investigates the feasibility and accuracy of employing thermal imaging for real-time
pothole detection, addressing the challenges associated with traditional methods. The methodology aims
to develop a real-time pothole detection system using thermal images through a deep learning approach,
specifically employing a CNN. The process begins with data acquisition using a FLIR ONE thermal
camera, capturing images of roads in Chandigarh city in India with and without potholes. Data is
collected based on pothole severity, shade presence, water presence, and various times of day and year.
Subsequently, data pre-processing involves cropping images to remove camera labels and resizing for
efficient processing. Augmentation techniques such as zooming, rotation, mirroring, and others are
applied to enhance the dataset [1].

The CNN model development includes a self-built CNN model and exploration of transfer learning with
pre-trained Residual Networks (ResNet) models. The self-built CNN model is a sequential architecture
with batch normalization, convolution layers, max pooling, global average pooling, and a dense layer,
utilizing ReLU activation, binary cross-entropy loss, and the Adam optimizer. Pre-trained ResNet
models are fine-tuned on the thermal image dataset, incorporating cyclic learning rates and differential
learning rates[1].

Advantages of the methodology include achieving high accuracy (97.08%) with ResNet and the
utilization of thermal imaging, providing benefits such as accuracy, low cost, and effectiveness in night
and fog conditions. The study recommends an optimal image dimension of 224x224 (the standard
ImageNet size) and highlights the positive impact of cyclic learning rates on accuracy improvement.
Drawbacks include the lower accuracy of the self-built CNN model and out-of-memory errors with
larger image sizes in ResNet models.

The proposed system demonstrates efficient pothole detection, showcasing its potential for accurate,

9
cost-effective, and versatile pothole detection in various conditions. Further enhancements, such as
detecting pothole severity, could further improve road maintenance efforts.The comparison with
previous works acknowledges the novelty of using thermal imaging for pothole detection, making direct
comparisons challenging due to variations in datasets and methodologies. The proposed technique,
however, stands out with its high accuracy and the application of thermal imaging. In conclusion, the
research contributes significantly to the field of pothole detection by introducing thermal imaging and
demonstrating the effectiveness of CNN-based ResNet models. The system's potential for real-time
implementation and its advantages over traditional methods underscore its importance in road
maintenance efforts. The suggestion for future work, including the detection of pothole severity and
region, further extends the applicability of the proposed system, paving the way for more advanced road
maintenance strategies.

The research works discussed above on automated pavement distress detection methods employ diverse
methodologies, each with its strengths and considerations. Koch and Brilakis (2011) focus on pothole
detection, utilizing a network of passenger vehicles equipped with technology for real-time assessment.
The two-stage video processing approach, image segmentation, and pattern recognition contribute to an
86% overall accuracy. Their systematic methodology, real-time defect identification, and transparency
about challenges enhance credibility. Yousaf et al. (2018) present an innovative top-down methodology
utilizing Bag of Words and SIFT features, achieving high accuracy (95.7%) and introducing a novel
localization index. The methodology's independence from assumptions about pothole shapes and
innovative evaluation criteria contribute significantly to asphalt pavement image categorization. Saad
and Tahar (2019) explore UAV technology for rut and pothole detection, showcasing detailed
measurements. The methodology lacks specific technical details but suggests potential with future work
on LiDAR UAV technology. Huidrom et al. (2013) propose a resource-saving methodology using video
distress frames selection and a critical distress detection algorithm, achieving impressive overall
accuracy rates of 97%. Challenges include a focus on a limited distress set and potential difficulties in
distinguishing distresses. Aparna et al. (2022) introduce thermal imaging and a CNN-based ResNet
model for real-time pothole detection, achieving 97.08% accuracy. The self-built CNN model and out-
of-memory errors in ResNet models pose challenges. While acknowledging drawbacks, their emphasis
on real-time implementation and future work on pothole severity detection opens avenues for advanced
road maintenance strategies. Each methodology contributes significantly, with future directions pointing
towards enhanced techniques, exploration of additional distress features, and advancements in sensor
technologies. Combining the strengths of these methodologies could lead to more robust and
comprehensive approaches for automated pavement distress detection in the future, utilizing various
technologies and innovative methodologies.

5. Conclusion

In conclusion, this literature review delves into the critical realm of Automated Pavement Distress
Detection, focusing on the imperative need for innovation in road maintenance management systems,
particularly with respect to pothole detection. The existing reliance on manual inspection methods is
highlighted as time-consuming and financially burdensome, emphasizing the urgency for automated
solutions. The review categorizes current research efforts into 3D reconstruction-based, vibration-based,
and vision-based methods, exploring their merits and challenges.

10
The urgency for automation is underscored by the pivotal role well-maintained asphalt pavements play
in fostering mobility and contributing to economic growth. Various distresses on pavement surfaces,
particularly potholes, are identified as significant concerns requiring comprehensive evaluation. The
conventional approach to distress data collection, predominantly manual, is discussed, revealing its
limitations in terms of expenses, time, and efficiency.

The literature review thoroughly examines the current advancements and challenges in pavement
distress detection. Notable research works, such as Koch and Brilakis (2011), Yousaf et al. (2018), Saad
and Tahar (2019), Huidrom et al. (2013), and Aparna et al. (2022), are critically analyzed. Each work
presents innovative methodologies ranging from networked passenger vehicles for real-time assessment,
hierarchical top-down approaches, UAV technology applications, video distress frames selection
algorithms, to thermal imaging with CNN-based models.

The strengths and considerations of each methodology are discussed, acknowledging advancements in
accuracy, speed, resource savings, and real-time implementation. Challenges such as equipment costs,
computational demands, and limitations in distress type coverage are recognized. The synthesis of these
methodologies is proposed as a pathway to more robust and comprehensive approaches for automated
pavement distress detection in the future, utilizing various technologies and innovative methodologies.

In essence, this literature review calls for continued refinement and innovation in automated distress
detection to enhance road maintenance practices. The research works examined provide valuable
insights and pave the way for future advancements, emphasizing the importance of addressing current
shortcomings for the evolution of efficient and reliable pavement distress detection methodologies.

5. References

[1] Aparna, Bhatia, Y., Rai, R., Gupta, V., Aggarwal, N., & Akula, A. (2022, March). Convolutional
neural networks based potholes detection using thermal imaging. Journal of King Saud University -
Computer and Information Sciences, 34(3), 578–588.
https://doi.org/10.1016/j.jksuci.2019.02.004

[2] Huidrom, L., Das, L. K., & Sud, S. (2013, December). Method for Automated Assessment of
Potholes, Cracks and Patches from Road Surface Video Clips. Procedia - Social and Behavioral
Sciences, 104, 312–321.
https://doi.org/10.1016/j.sbspro.2013.11.124

[3] Koch, C., & Brilakis, I. (2011, August). Pothole detection in asphalt pavement images. Advanced
Engineering Informatics, 25(3), 507–515.
https://doi.org/10.1016/j.aei.2011.01.002

[4] Saad, A. M., & Tahar, K. N. (2019, April). Identification of rut and pothole by using multirotor
unmanned aerial vehicle (UAV). Measurement, 137, 647–654.
https://doi.org/10.1016/j.measurement.2019.01.093

11
[5] Yousaf, M. H., Azhar, K., Murtaza, F., & Hussain, F. (2018, October). Visual analysis of asphalt
pavement for detection and localization of potholes. Advanced Engineering Informatics, 38, 527–537.
https://doi.org/10.1016/j.aei.2018.09.002

12