1

Automated Pothole and Road Irregularity Detection

for Driver Safety
Dr. H. Devanna Appikonda Tanush Kumar
Ambati Karthik
Department of Electronics and Department of Electronics and
Department of Electronics and
Communication Engineering Communication Engineering
Communication Engineering
Vignan’s Institute of Information Vignan’s Institute of Information
Vignan’s Institute of Information
Technology(A) Technology(A)
Technology(A)
Duvvada, Vizag Duvvada, Vizag
Duvvada, Vizag
hdevanna@gmail.com appikondatanush9989@gmail.com
ambatikarthik2003@gmail.com

D.K.V. Varma G.S.L. Manasa

Department of Electronics and Department of Electronics and
Communication Engineering Communication Engineering
Vignan’s Institute of Information Vignan’s Institute of Information
Technology(A) Technology(A)
Duvvada, Vizag Duvvada, Vizag
datlakrishnavivekvarma@gmail.com manasagamini2004@gmail.com

This results in a cycle of reactive repairs, where potholes are

Abstract— Roadside potholes can cause severe traffic patched after they’ve already caused damage or posed a safety
safety issues and damage vehicles. Pothole detection is an risk. Approximately drivers have spent over Rs. 10,000 crores
important activity in road maintenance that ensures traffic in five to seven years on automobiles to repair damage caused
safety and infrastructure longevity. This work proposes an by potholes. Each motorist incurred a cost of almost Rs.
effective method for pothole detection based on adaptive
10,000 on an average. The survey in India states that around
Canny edge detection and contour analysis. The technique
5,000 accidents on an average takes place. So, vehicles need to
dynamically calculates edge thresholds according to the
get upgraded by pothole detection model so it lowers it speed
median intensity of the image, providing robustness in
different lighting conditions. The pipeline starts with pre- or stops [1].
processing with Median Blur to minimize noise, and We believe that technology can play a crucial role in
adaptive Canny edge detection for reliable boundary breaking this cycle and ushering in a new era of intelligent
detection. Extracted edges are then detected using contour infrastructure management. Over the past decade, various
detection, where convex hull methods and bounding box technologies such as active and passive sensing, have been
computations assist in marking potential pothole areas. To utilized to acquire road data and aid personnel in detecting
enhance reliability, DBSCAN clustering clusters proximate road potholes [2]. These methods include using vibration
contours to reduce false detection. The new method analysis from 3-axis accelerometer [3], disparity
improves conventional edge detection by learning to adapt transformation and road surface modeling [2], computer vision
to various environments and minimizing manual threshold and learning [1], mobile laser scanning [4], point laser sensor
tuning. The efficiency of the system makes it ideal for [5], pot-yolo v8 network [6]. To overcome this, we try to
practical applications like autonomous road inspection and propose a road irregularity detection technique using
self-driving vehicle navigation, with a cost-efficient and Raspberry Pi and image processing combined with GPS for
scalable approach to pothole detection. location tracking.
Index Terms—Raspberry Pi, Canny edge detection, Cost-
effective solution, Road safety. II. LITERATURE REVIEW

I. INTRODUCTION A. Ultrasonic GPS Mapping

Potholes. They are more than just a nuisance; they’re a To find the depth of any potholes, the pothole
symptom of a complex interplay of factors: aging detection system uses a waterproof ultrasonic sensor
infrastructure, increasing traffic volume, budget constraints that is attached on the front end of the Pothole Robot
and often a reactive approach to maintenance. Agent (PRA). The sensor constantly senses the
We live in a world where technology permeates nearly every distance on a flat road. A pothole is a positive
aspect of our lives, yet road maintenance often lags behind. difference, but a negative difference indicates a road
Traditional methods of pothole detection – relying on citizen hump. To enable correct mapping of pothole
complaints, infrequent visual inspections, or even just waiting locations, the system also utilizes a GPS Module
Board to take the geographic coordinates of sensed
for the damage to become severe enough to warrant attention
potholes. An Atmega328 Arduino Uno
– are simply not efficient or proactive enough.
Microcontroller manages the communication
 2

between the GPS module and the ultrasonic sensor, E. Mobile laser scanning
guaranteeing precise data processing and gathering. It is a semi-automated method for classifying road
This process facilitates easy detection and definition markings from Mobile laser scanning point clouds.
of potholes in real time, which later forms an The process involves steps such as Preprocessing,
extensive pothole map for road users. [7] Road Marking Extraction and Classification. In
preprocessing, raw 3D MLS data is filtered to remove
B. POT-YOLO objects other than road using voxel-based
With the YOLOv8 network, the authors introduced a segmentation and high-pass filtering, then converted
novel approach for real-time pothole detection into 2D intensity images with scan angle based
termed POT-YOLO. The authors captured pothole correction to enhance contrast. Next, Otsu’s
videos and split them into individual picture frames thresholding and noise removal techniques extract
for further analysis. They employed a pre-processing road markings and they are segmented and classified
technique referred to as the Contrast Stretching using geometric features and hierarchical
Adaptive Gaussian Star Filter (CAGF) to enhance the decision tree. [4]
quality of these frames and eliminate distortions.
Following pre-processing, Sobel edge detection was III. METHODOLOGY
applied to scan the photos and identify the edges of A. Block Diagram
potholes. To identify and classify various types of
potholes such as cracks, oil spots, patches, and stones The proposed pothole and road irregularity detection
correctly, the processed photos were then input into system in Fig. 1 consists of multiple hardware and
the POT-YOLOv8 model. Following several metrics, software components, which work together to
the performance of the POT-YOLO method was capture, process, store and inform users about
measured and revealed to be highly accurate and potholes and irregularities.
efficient in pothole detection but is highly
complicated. [6]

C. Vibrational Analysis
It employs vibration sensors attached to a vehicle's
steering column to sense road roughness. The data
are measured using a triaxial accelerometer,
positioning using GNSS, and an NB-IoT module for
wireless communication. The vibration signals are
analyzed with the use of RMS calculations, FFT, and
wavelet transforms in order to detect potholes, 80 HZ
vibrations being signals generated from extreme road Fig. 1: Block diagram of proposed detection system
damage. The system maps potholes by merging
vibration signals with location data using Google In our project, a camera module is mounted on a vehicle and
Plus codes for precise tracking. Data is passed on to a continuously captures road images, which are sent to the
cloud-based Web-GIS environment that offers real- Raspberry pi board making it the primary input source, the
time updates of road conditions, but is unable to offer GPS module provides real time location data which helps to
data from the path the car has not traveled, or when map the detected potholes with their exact geographic
the pothole is found in the center of track. [3] coordinates and the raspberry pi board acts as the processing
unit of the system. It runs computer vision algorithms using
D. Point Laser Sensor OpenCV to detect potholes from captured images, it also
The authors presented a real-time road assessment handles image segmentation, classification and object
system using point laser sensors and a low- detection. The images are processed in real time which include
complexity signal processing algorithm to detect road multiple steps like preprocessing, noise reduction and contour
surface defects. Unlike traditional scanning lasers, detection. The system applies segmentation techniques to
which provide high-density 2D data, the proposed separate potholes from the background, which helps in
system captures longitudinal 1D data, reducing improving the accuracy. After detecting it classifies the
complexity and enabling real-time processing. The pothole based on the severity and size and stores the images in
method calculates a Road Texture Index (RTI) using a cloud database, so the authorities can access the data for
Root Mean Square (RMS) values and compares local road maintenance planning. Further, it alerts the driver by
and global RTI distributions to identify road sending a message through telegram bot.
deterioration.
Performance comparisons with scanning lasers show B. Flowchart
that point lasers effectively identify longitudinal
defects but may miss small traverse cracks. [5]
 3

computer vision library containing a suite of tools for image

analysis and processing. The use of OpenCV is necessary for
the execution of tasks like edge detection, contour analysis,
and filtering of images, which are indispensable in the
detection of potholes in taken photographs. For the purpose of
promoting user interaction and real-time feedback, the project
incorporated the Telegram Bot API. This API enables the
system to notify the users whenever a pothole is detected.
Through the development of a bot using the Telegram
application, developers can receive a unique token that
identifies the bot and allows it to send messages to given chat
IDs. The notifications contain critical information, such as the
GPS coordinates of the pothole that has been detected as well
as the number of potholes detected, so that users are notified
instantly of any problem that needs to be addressed. This
feature greatly enhances the system's responsiveness so that
users can take instantaneous actions based on the notifications
received. The Dropbox API was used in combination with
cloud storage in the project. This enables the uploading of
images captured to Dropbox, giving a safe and accessible
place for remote storage of images. Through the use of cloud
storage, the system guarantees that images are saved and are
also readily accessed from any internet-enabled device. This
assists in giving the information to local authorities for road
Fig. 2: Flowchart maintenance and repairing them to minimize the accidents.

IV. MATHEMATICAL CALCULATIONS

1. Conversion to Grayscale: The grayscale conversion is

performed to simplify the computation and improve
the edge detection and also to reduce the memory
usage. The conversion is performed through a
weighted sum of red (R), green (G), and blue (B)
channels.
𝑌 = 0.299 ⋅ 𝑅 + 0.587 ⋅ 𝐺 + 0.114 ⋅ 𝐵

2. Histogram Equalization: Histogram Equalization is

used to enhance the contrast of the image. It achieves
this by distributing the highest occurring intensity
Fig. 3: Components and their functions values, so that dark areas are made lighter and light
areas more textured.
The project is constructed around Raspberry Pi, a tiny, low- 𝑠𝑘 = (𝐿 − 1) ⋅ 𝐶𝐷𝐹(𝑟𝑘 )
cost computer that is extremely versatile and can run many Where L is the maximum intensity level and rk
different applications. Raspberry Pi acts as the core processing represents the original intensity of image before
unit of the whole system, allowing it to take images, process equalization, and The CDF makes sure that the
information, and talk to external services. The Raspberry Pi frequent intensities are distributed more.
Camera module was used for image capture. This camera is
tailored for use with Raspberry Pi and is capable of high- 3. Median Blur: Median blurring is a nonlinear filter
resolution image, normally up to 1080p. The capability of the used to attenuate noise without losing edges of the
camera to be able to capture clear images is important in image. It replaces the value of each pixel with the
image-processing applications, in pothole detection. Besides, median value of pixels within its neighborhood.
𝐼𝑚𝑒𝑑𝑖𝑎𝑛 (𝑥, 𝑦) = 𝑚𝑒𝑑𝑖𝑎𝑛{𝐼(𝑖, 𝑗) (𝑖, 𝑗) ∈ 𝑊𝑖𝑛𝑑𝑜𝑤𝑘×𝑘 (𝑥, 𝑦)}
an GPS module, NEO-6M, will be incorporated into the
system. This module offers real-time GPS coordinates such 𝑊𝑖𝑛𝑑𝑜𝑤𝑘×𝑘 (𝑥, 𝑦) is the collection of pixels in a
that the system can record the precise locations of potholes 𝑘 × 𝑘 window centered at (𝑥, 𝑦)and the median is
that have been detected, which is very important in mapping. the middle value of the sorted intensities in the
The development environment for this project was mostly window.
built around Python, a strong and popular programming 4. Adaptive Canny Edge Detection: Adaptive Canny
language that is noted for its readability and robust library Edge Detection automatically adapts the lower and
support. The code makes use of OpenCV, an open-source upper threshold values for edge detection rather than
 4

employing fixed values. It supports the processing of

different lighting conditions and contrast levels in
images. The lower threshold is calculated as,
𝑙𝑜𝑤𝑒𝑟 = 𝑚𝑎𝑥(0, (1 − 𝜎) × 𝑣)
The upper threshold is calculated as,
𝑢𝑝𝑝𝑒𝑟 = 𝑚𝑖𝑛(255, (1 + 𝜎) × 𝑣)
Where, 𝜎 is the sensitivity range which ranges
between 0.1 to 0.5, where 0.1 is taken for noisy
images and 0.5 is taken for low-contrast images, and
𝑣 is the median value of pixel intensities of the input
image. It is used as a point of reference to Fig. 5: Blurred image
automatically select optimal edge detection Then, the image goes through adaptive canny edge detection,
thresholds which automatically adapts the lower and upper threshold
values to detect the edges in the images, all the edges with
threshold above the upper threshold or below the lower
V. RESULTS AND DISCUSSION
threshold are avoided. It allows images with different lighting
conditions and contrast levels.
The project rendered outstanding outcomes, highlighting an
uncomplicated marriage of technology for fortifying urban
infrastructure management. Raspberry Pi was used
successfully to identify high-resolution pictures of potholes,
boasting a 70-85% accuracy of detection via image-processing
algorithms that discerned real threats from other roadside
anomalies. Upon detection, the system sent a message
instantly to a Telegram bot, so the users would obtain real-
time reports and response can be triggered instantaneously by
the maintenance crew. This new communication channel not
only informed people but also provided a platform through
which users could report other potholes they came across. Fig. 6: Edge detection
In addition, the project utilizes an Excel sheet to save data, The contours are detected in the edge detected images as it
such as the location, size, images of potholes detected. This converts disconnected edge pixels into structured, meaningful
organized data collection compels city planners to focus on objects and extracts the boundaries for better detection.
repairs by severity and frequency. In general, this project not
only shows the strength of low-cost technology but also
highlights the importance in creating safer, smoother roads for
everyone.
The detection uses various image processing algorithms for
preprocessing the images to remove any noise without
affecting the image like, we covert the original image to
grayscale to simplify the computation and reduce the memory
usage.

Fig. 7: Detected contours

Fig. 4: Grayscale image

The grayscale image undergoes through median blur
technique, which attenuates the noise without losing edges of
the image. It replaces the value of each pixel with the median Fig. 8: Circuit connections
value of pixels within its neighborhood.
 5

2. Extended Road Anomaly Detection:

Extend the system to identify other roads anomalies
like cracks, speed bumps, and debris, offering a more
holistic road condition monitoring solution.

3. Real Time notifications through Mobile Apps:

Create a dedicated mobile app to send real-time alerts
to drivers in the form of visual and audio signals and
recommend alternative routes to bypass potholes
Fig. 9: Detected pothole .
4. Integration with Traffic Management Systems:
Integrate the system with traffic management centers
to make dynamic changes in traffic signals or speed
limits for zones with road hazards detected,
enhancing overall traffic safety.

ACKNOWLEDGMENT
We further thank Vignan's Institute of Information
Fig. 10: Output data stored in an excel sheet Technology for their collaboration during the tenure of this
research project.

REFERENCES
[1] Bidve, Vijaykumar S., et al. "Pothole detection model for road
safety using computer vision and machine learning." Int J Artif
Intell ISSN 2252.8938: 4481.
[2] Fan, Rui, et al. "Pothole detection based on disparity
transformation and road surface modeling." IEEE Transactions
on Image Processing 29 (2019): 897-908.
[3] Chen, Dong, et al. "Real-time road pothole mapping based on
vibration analysis in smart city." IEEE Journal of selected topics
in applied Earth observations and remote sensing 15 (2022):
6972-6984
[4] Cheng, Ming, et al. "Extraction and classification of road
markings using mobile laser scanning point clouds." IEEE
Fig. 11: Alerts sent to user through Telegram Journal of Selected Topics in Applied Earth Observations and
Remote Sensing 10.3 (2016): 1182-1196.
VI. CONCLUSION AND FUTURE SCOPE [5] Li, Wenda, Michael Burrow, and Zijun Li. "Automatic road
condition assessment by using point laser sensor." 2018 IEEE
SENSORS. IEEE, 2018.
In summary, we feel that this project revolutionizes how we [6] Bhavana, N., et al. "POT-YOLO: Real-Time Road Potholes
address road hazards using the prowess of technology. Thanks Detection using Edge Segmentation based Yolo V8 Network."
to real-time detection, this new solution makes the roads safer IEEE Sensors Journal (2024).
and enables communities to play an active role in reporting [7] Z. Hou, K. C. P. Wang, and W. Gong, “Experimentation of 3D
problems. Past studies that attained promising pothole pavement imaging through stereovision,” in International
detection need more expensive and sophisticated Conference on Transportation Engineering 2007, ICTE 2007,
environments than the solution outlined in this present work. 2007, pp. 376–381, doi: 10.1061/40932(246)62.
[7]-[10] By opening the door to intelligent urban [8] H. Lokeshwor, L. K. Das, and S. Goel, “Robust method for
infrastructure, this project demonstrates how a humble automated segmentation of frames with/without distress from
road surface video clips,” J. Transp. Eng., vol. 140, no. 1, pp. 31–
Raspberry pi can have a profound effect on our everyday lives, 41, Jan. 2014, doi: 10.1061/(ASCE)TE.1943-5436.0000564.
transforming rough roads into smooth and secure rides. In the [9] A. Mednis, G. Strazdins, R. Zviedris, G. Kanonirs, and L. Selavo,
future the project can be extended further to various ways like: “Real time pothole detection using Android smartphones with
accelerometers,” in 2011 International Conference on Distributed
1. Integration with Self-Driving Cars: Computing in Sensor Systems and Workshops, DCOSS’11, 2011,
The system can be modified to be used in self-driving doi: 10.1109/DCOSS.2011.5982206.
cars to improve their navigation and safety [10] R. S. Shaikh Patil Pallavi and S. S. Priyanka Khan Aabid A,
capabilities by offering real-time pothole detection “Detection of Potholes, Humps and Measuring Distance between
and avoidance features. Two Vehicles using Ultrasonic Sensor and Accelerometer,” GRD
Journals-Global Res. Dev. J. Eng., vol. 3, no. 6, 2018.