Volume 9, Issue 11, November – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24NOV509

Pothole Detection and Geological

Mapping through Aerial Vehicle
Anushka Kadam1; Anushka Patil2; Aditya Pore3
Department of Electronics and Telecommunication
COEP Technological University
Wellesly Rd, Shivajinagar, Pune, Maharashtra 411005

Abstract:- The construction and development of good

roads is important for the economy and prosperity of
today's countries. Road maintenance faced many
challenges due to heavy traffic, insufficient funds, and
lack of resources. These roads and highways are very
dangerous for drivers due to damage such as potholes.
Bad roads can lead to dangerous problems such as
damaged wheels, tires, damaged vehicles, serious
accidents, irregular traffic, and poor driving. To better
solve this problem, this project presents a deep learning
method that can identify and classify pots according to
their size and severity. Drones, also known as unmanned
aerial vehicles (UAVs), can be utilized to modernize road
safety practices and so enhance the nation's Fig 1: Potholes
infrastructure. The image library is
employed for the development of a pothole detection The existing pothole detection systems are deficient in
model, which is subsequently utilized within an some areas, such as high accuracy rates or a large need for
algorithmic framework that integrates a road color computing resources, which is not practical everywhere in
model with fundamental image processing techniques, the world. Tens of thousands of deaths are attributed to
including a Canny filter and contour detection. potholes every year across the world. The ability to detect
This approach enhances the level of accuracy in identify these potholes using image processing enables us to identify
ing potholes. and categorize them based on their condition and the level of
harm they possess. Such a serious situation calls for a faster
Keywords:- Pothole, Unmanned Aerial Vehicles (Uavs), response. I was motivated to develop an algorithm that can
YOLO V8, Geotagging and Geomapping. accurately identify them and warn users of potholes that pose
a higher risk. The goal in creating this algorithm was to help
I. INTRODUCTION save lives and stop traffic accidents.

Potholes are 10-inch-deep, bowl-shaped gaps in the II. LITERATURE REVIEW

road brought on by excessive wear and ageing of the
pavement. They can be seen growing on the top layer of a To detect potholes, Buza E. [1], Seung-KiRyu [2],
road, which typically deteriorates from weathering, heavy Christian Koch [3], and Manisha Mandal [4] employed a
load, bearing trucks, or normal wear and tear. One of the segmentation method based on a histogram and clever edge
most frequent causes of vehicle accidents is a pothole, which detection. To identify potholes, many researchers most
can deepen to several inches with rain. They frequently frequently employ the edge detection and thresholding
cause auto accidents, but they can also be fatal for techniques. Vignesh War K et al. [5] categorized potholes
motorcycles. Road potholes are extremely dangerous while using a variety of methods, including edge detection, image
travelling at high speeds since they are difficult to see from segmentation, and clustering methods including K-Means,
a distance. thresholding, and fuzzy C-Means. Instead, then creating a
stand-alone product, this article primarily compares several
Two-dimensional (2D) vision-based) methods can also image segmentation algorithms. According to Shambhu
be used to find potholes. The evaluated shape of the pothole Hegde et al. [6], an ultrasonic sensor that measures pothole
is decided employing a square tile matrix to speak to depth is installed on the car. A microcontroller uses and
pothole-related locales. The 2-dimensional perception processes the sensor's output before transmitting the data to
method is limited to uniform lighting conditions and is other cars. Potholes are found using computed sensor
unable to gauge the precise depth of potholes. To get over readings; no specific classification process is employed.
the drawbacks of earlier systems, a 2-dimensional parameter Vinay Rishiwal et. al.[7] made advantage of an
checking method based on computer stereo vision issued. accelerometer that is already present in an android handset.

IJISRT24NOV509 www.ijisrt.com 1200

Volume 9, Issue 11, November – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24NOV509

After that, it was built, and the axis acceleration data was the pothole area was added to the image in order to correctly
obtained. The z-axis acceleration, which experiences abrupt partition the pothole zone.
changes when a pothole is met, is monitored in order to
identify potholes. To detect potholes, Kiran Kumar et al. [8] III. PROPOSED METHODOLOGY
equipped the car with a laser and camera setup. The camera
takes an image of the reflected laser and uses it as a The proposed study aims to create a drone-based
reference. Given that it is a physics-based approach and pothole detection system. It helps the relevant authorities
requires intricate mathematical calculations, it might not be identify and address the problem. The pothole detecting
very accurate. In Danti et al. (2012), the primary method for technology will be installed on a drone that is either
detection and classification was an image processing manually or automatically piloted. A "Camera" is used as a
algorithm-based system. The study sought to identify several sensor to check the road for potholes. The main layout of our
traffic-related items, including lanes, road signs, and model is shown in Figure 2.
potholes. A black and white threshold that would emphasize

Fig 2: Block Diagram

A. Data Collection and Preparation avoiding over-fitting. For augmentation, a number of factors
There are just a few officially sanctioned publicly were employed, including scale, color modifications and
available datasets for pothole identification and no online rotation.
potholes dataset. We thus gathered pothole photographs for
our study from many sources, including Kaggle and C. Image Annotation
Roboflow, and we utilized drone to gather more images from Labels were applied to the photos that were acquired
roadways.80 photos were used to test the model that was using image annotation. A human operator examines a
produced, 183 images were used to validate the model, and collection of photos and spots any pertinent potholes. To
450 images were utilized for training. The dataset's photos construct a training dataset for computer vision models,
contain a variety of potholes with various sizes, areas, and these annotations can be employed. The model learns to
depths. Our PASCAL/VOC-formatted photos were labelled recognize items or label photos on its own using human
using CVAT, an open-source graphical annotation tool, annotations as its ground truth. Models for tasks including
before being converted to.txt format for YOLOv8. image classification, object identification, and picture
segmentation may be trained using this procedure. We have
B. Image Augmentation and Data Formatting annotated oven 450 images using CVAT tool. CVAT is a
Over-fitting may result from the dataset's small size. online tool for annotation. We can annotate in boxes,
Deep learning models, however, need a sufficient volume of polygon shapes. Then the data is exported in the format of
data to provide reliable findings. As a result, we have used yolo1.1. The annotated images having text files followed by
augmentation strategies to benefit from regularization while their bounding box coordinates as shown in Figure 3.

IJISRT24NOV509 www.ijisrt.com 1201

Volume 9, Issue 11, November – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24NOV509

With 70 epochs, the total workout lasts around 45 minutes.

With epochs, the mAP@50 and mAP@50-95 are
dramatically rising. Class loss and box loss both declines.
The model is stored with the Python extension (.pt) during
execution.

E. Geo Tagging and Geo Mapping

The program receives pictures taken by drones. Python
was used to create the program. It extracts coordinates in
latitude and longitude from picture information using the
exifread library.

IV. RESULT AND DISCUSSION

Using the training and testing data acquired from the

Fig 3: Image Annotation tagged photographs, the performance of the YOLOv8 model
in terms of recognizing and categorizing potholes was
D. YoloV8 Model Training evaluated. After the algorithm has been trained and
The Colab runtime notebook is used to complete the validated, it is ready to be tested against datapoints of
training. The COLAB runtime has been loaded with the different pothole severity levels. It can successfully classify
whole dataset. The dataset includes the text files for the them and determine whether work needs to be done. The
photos. 3.8 Python is employed. It uses the Yolov8n model. YOLOv8 may produce results with high accuracy of 90% or
We are using Python to copy the Yolov8n model from the more.
Ultralytics package. There were 70 training epochs. It is split
the validation dataset. Tesla K80 GPU is the one being used. Following is some sample predictions on validation
dataset.

Fig 4: Detected Pothole Samples

IJISRT24NOV509 www.ijisrt.com 1202

Volume 9, Issue 11, November – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24NOV509

Table 1: Training Results for yolo v8 Model

Sr. No Parameters Patterns
1 Precision curve(p-curve)

2 RecallCurve(R-curve)

3 PR Curve

4 F1 Curve

A. WebApp for user Interface users to report. The most important task was creating an
The website has been created for detecting and algorithm for detecting potholes from image and video
geotagging potholes to help the user to identify and repair footage. The Web App image is mentioned in fig 5.
potholes in a timely manner. We started by planning the
website and designing the user interface to make it easy for

IJISRT24NOV509 www.ijisrt.com 1203

Volume 9, Issue 11, November – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24NOV509

Fig 5: Website to Detect Potholes Fig 7: YOLO v8 Output

B. WebApp for user Interface V. FUTURE SCOPE AND CONCLUSION

By using the coordinates, it becomes easy to locate
potholes on google maps through Geo mapping. Fig 6 shows It's critical to recognize the potential of this developing
the detected potholes on the maps for quick access to the sector for enhancing road maintenance and safety as it relates
location. to the future use of pothole identification using satellite
photography. It may be able to create systems that can
monitor roads in real-time and find potholes as soon as they
appear to the expanding accessibility of satellite images.
Another option is to equip drones with sophisticated sensors
in addition to their cameras, such LiDAR or radar. These
sensors can offer additional information on the road surface,
such as the size and depth of potholes, which can aid in
determining the priority of maintenance and repairs. A
website that is connected with the government allows
individuals to contribute information about the state of the
roads in their area, and the government receives the
coordinates to locate the area where potholes are likely to
appear quickly.

In conclusion, the monitoring and mapping system

for potholes is deemed essential in averting traffic accidents
and ensuring enhanced roadconditions nationwide, thereby
fostering accelerated development.

REFERENCES

[1]. Vigneshwar K, Hema Kumar B, 2016 IEEE

Conference on Computational Intelligence and
Fig 6: Location on Google Maps Computing Research. 08 May 2017, 2473-943X
[2]. Shambhu Hegde , Harish V, Mekhali ,
C. Detected Pothole using YOLO GollaVaraprasad, Pothole Detection and Inter
The Model which was trained and tested can be vehicular Communication 2014 IEEE International
deployed to detect the potholes in new images and videos as Conference on Vehicular Electronics and Safety.
978-1-4799-1882-9. 23. March 2015. 84 -87.
shown in figure 7.
[3]. Vinay Rishiwal and Hamshan Khan, Automatic
Pothole and sped breaker detection using Android
System . 2016 MIPRO June 2016. Pp1270-1273.

IJISRT24NOV509 www.ijisrt.com 1204

Volume 9, Issue 11, November – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24NOV509

[4]. Kiran Kumar Vupparaboina, Roopak R Tamboli ,

P.M Shenu and Soumya Jana. Laser based detection
and depth estimation of Dry and Water Filled
Potholes , A geometric approach. 2015 NCC (2015)
March
[5]. Kanza Azhar, Fiza Murtaza, Muhammad Haroon
Yousaf and Hafiz Adnan Habib, Computer Vision
based detection and localization of potholes . IEEE
CCECE 2016 May
[6]. Vosco Periera, Satoshi Tamura, Satoru Hayamizu
and Hidezaku Fuka. A Deep learning based approach
for Road Pothole Detection in Timor Leste 2018
IEEE SOLI(2018) Aug. pp275-284
[7]. S Rode, S Vijay, P Goyal, P Kulkarni, K Arya.
Pothole Detection and Warning System:
Infrastructure support and System Design,
International Conference on Electronic Computer
technology. (2009) Feb pp 286-290
[8]. Byeong-ho Kang, Su-li Choi. Pothole Detection
system using 2D LiDAR camera. 2017 ICUFN. July
pp744-746.
[9]. Ionut Schiopu, Jukka P. Saarinen, Lauri Kettunen,
Ion Tabus. Pothole detection and tracking in car video
sequence. 2016 TSP. June pp701-706.

S Rode, S Vijay, P Goyal , P Kulkarni, K Arya. Pothole Detection and Warning System: Infrastructure support and System Design, International Conference on

IJISRT24NOV509 www.ijisrt.com 1205