Volume 8, Issue 2, February – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Android Pothole Detection System

Using Deep Learning

Pranav Prashant Kulkarni Ketan Mandar Kulkarni

Dept. of Information Technology Dept. of Information Technology
Sinhgad College of Engineering Sinhgad College of Engineering
Vadgaon BK, Pune, Maharashtra, India Vadgaon BK, Pune, Maharashtra, India

Vivek Rajkumar Nimbalkar S. P. Jadhav

Dept. of Information Technology Professor, Dept. of Information Technology
Sinhgad College of Engineering Sinhgad College of Engineering
Vadgaon BK, Pune, Maharashtra, India Vadgaon BK, Pune, Maharashtra, India

Abstract:- In recent years, technology has contributed to I. INTRODUCTION

improving transportation systems, but the maintenance
and upkeep of road networks remain a challenge despite In recent years, there have been numerous advancements
these advancements. Potholes, cracks, and other road in technology that have contributed to the improvement of
defects can lead to accidents, traffic congestion, and transportation systems. For instance, the use of GPS and
expensive repairs. An android pothole detection system sensors in vehicles and on roads has greatly enhanced the
that utilizes smartphone sensors and machine learning efficiency and safety of transportation. Additionally, the
algorithms has the potential to revolutionize road development of autonomous vehicles has the potential to
maintenance and safety. This proposed system will use a revolutionize the transportation industry, providing safer and
smartphone camera to detect potholes and distinguish more convenient means of travel.
them from other road irregularities. It will also be
integrated with a backend database that can store and However, despite these advancements, there are still
analyze data on road conditions, enabling authorities to many challenges that need to be addressed. One of the biggest
prioritize maintenance and repairs. The android pothole challenges is the maintenance and upkeep of road networks,
detection system can offer several benefits, including particularly in areas with high traffic volumes. Potholes,
reducing road accidents, lowering repair costs, and cracks, and other road defects are common occurrences, and if
minimizing traffic congestion. Our proposed solution aims left unaddressed, they can lead to accidents, traffic congestion,
to crowdsource information from people who face these and costly repairs.
issues and forward it to relevant authorities using an
Android application. To achieve this, we will utilize a Deep The implementation of a pothole detection system can
Learning model capable of detecting potholes, collecting significantly mitigate these issues. By detecting and reporting
information from users, and sending it to authorities. The road defects in real-time, authorities can take prompt action to
success of this solution depends on the accuracy of the repair and maintain road networks, preventing accidents and
Deep Learning model, the quality of user-provided other associated problems. Moreover, the system can provide
information, the responsiveness of relevant authorities, authorities with data on the frequency and severity of road
and user engagement. Therefore, the system must have defects, enabling them to make informed decisions on
appropriate parameters to manage these factors and prioritizing repairs and allocating resources.
guarantee the solution's effectiveness. In conclusion,
utilizing technology for android pothole detection can lead Overall, the development of a pothole detection system is
to effective and timely repairs, contributing to overall road crucial in ensuring the safety and functionality of road
safety and reducing vehicle damage. The proposed android networks. By leveraging technology, we can efficiently and
pothole detection system brings people together to work on effectively address road defects, promoting safe and efficient
a common problem and has the potential to revolutionize transportation for all.
road maintenance and safety, providing safer and more
efficient means of travel for everyone. II. BACKGROUND

Keywords:- Deep Learning , Road Safety , Efficient Means To Road networks are crucial for connecting communities
Travel , Smartphone Sensors. and supporting economic growth, but regular usage and other
factors can lead to potholes and other road defects that pose
safety risks and cause inconvenience for commuters.
Traditional methods of pothole detection can be time-
consuming and inaccurate, but leveraging technology such as
sensors and machine learning algorithms can provide real-time
and accurate data on pothole locations and severity. The

IJISRT23FEB1109 www.ijisrt.com 1482

Volume 8, Issue 2, February – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
development of an android pothole detection system that can profile and identify the location of pavement deformities. The
leverage smartphone sensors and machine learning algorithms system also takes into account the variations in accelerometer
has the potential to revolutionize road maintenance and safety. values in relation to the vehicle speed.
The proposed system will use the smartphone's GPS sensors to
detect potholes and distinguish between potholes and other Issue - Merely relying on an accelerometer does not ensure
road irregularities. The system will also be integrated with a accurate results, and this approach is ineffective when a person
backend database that can store and analyze data on road is not riding a vehicle.
conditions, allowing authorities to prioritize maintenance and
repairs. The android pothole detection system can provide Author - Pooja P.R & Balaji Hariharan
several benefits, such as reducing road accidents, lowering
repair costs, and minimizing traffic congestion. The  Report titled (2021) -Deep Learning Approach to Detect
development of such a system has the potential to revolutionize Potholes in Real-Time using Smartphone.
the transportation industry, providing safer and more efficient The identification and timely mapping of potholes is
means of travel for all. crucial to prevent road accidents. Currently, identifying
roadway damage is a labor-intensive manual process. This
III. LITERATURE SURVEY research proposes a system that utilizes deep learning
algorithms and integrates with smartphones to detect potholes
 Report titled (2020) -A modern pothole detection technique in real-time. The system includes a smartphone application that
using deep learning. maps all potholes along the user's route. Meanwhile, the deep
In India, road accident detection and avoidance pose a learning object detection algorithm, Single Shot Multi-box
significant challenge due to the use of poor quality construction Detector (SSD), runs in the background, using the mobile
materials in the road drainage system. These issues lead to early camera to detect potholes. Whenever an unregistered pothole is
road damage and the formation of potholes, which in turn can detected by SSD, the coordinates are updated in the database in
cause accidents. According to a report submitted by the real-time. In addition, a Deep Feed Forward Neural Network
Ministry of Road Transport and Highways transport research model continuously assesses accelerometer and gyroscope
wing in New Delhi, approximately 4,64,910 accidents occur readings to detect unregistered potholes. The dual mechanism
per year in India. To mitigate this problem, this paper proposes of camera-based and accelerometer-gyroscope based detection
a deep learning-based model that uses images and videos to not only cross-validates detections but also provides stable
detect potholes early, thereby reducing the chances of results even if one mechanism fails. The pothole coordinates
accidents. The model is based on Transfer Learning, Faster are rendered on the map user interface within the same
Region-based Convolutional Neural Network (F-RCNN), and application. With a map/navigation feature as the front end and
Inception-V2. While there are many models for pothole a two-fold deep learning pothole detection algorithm as the
detection that use machine learning techniques with backend, this system is an efficient and cost-effective solution
accelerometers, relatively few models can be found that rely for real-time pothole detection.
solely on machine learning techniques to detect potholes using
images and videos. The results of this study demonstrate that Issue - It is unnecessary to determine the area of individual
our proposed model outperforms other existing techniques for potholes since there are often numerous potholes on a road, and
pothole detection. this is an insufficient parameter. Contracts are typically issued
to repair the entire road to address this issue.
Issue - The model's performance is significantly impacted
when it is exposed to natural phenomena such as rainfall or Author - Shubham Kokate , Uday more
snowfall.
IV. METHODOLOGY
Author - Abhishek kumar
Our proposed solution aims to address ongoing road
 Report titled (2017) - An Early Warning System for Traffic problems by crowdsourcing information from people facing
and Road Safety Hazards Using Collaborative Crowd these issues and forwarding it to relevant authorities using an
Sourcing. Android application. To achieve this, we will utilize a Deep
The growing number of vehicles on the roads poses a Learning model capable of detecting potholes, collecting
challenge for the road maintenance department to address the information from users, and sending it to authorities. The steps
demand for timely repairs, necessitating the need for a involved in implementing this solution are depicted in the
comprehensive approach To detect road construction issues figure below.
such as potholes, bumps, corrugations, waves, and defective
street cuts.. However, without an improved real-time traffic
alerts system, efficiently maintaining city roads will be a
daunting task. Therefore, there is a pressing need for a more
effective monitoring system that can detect and resolve road
infrastructure issues. In this research, an analytical model of a
system is presented that can detect pavement deformities. The
system employs a mobile application that uses the in-built
accelerometer in smartphones to capture the accelerometer

IJISRT23FEB1109 www.ijisrt.com 1483

Volume 8, Issue 2, February – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
pothole images to test various object detection models. The
proposed architecture utilizes a smartphone camera to capture
images of potholes and sends sufficient information to
authorities for them to take action with priority. This approach
will not only provide drivers and commuters with advisories
for safe driving but also deal with local issues that arise. In
conclusion, utilizing such technologies for pothole detection
can lead to effective and timely repairs, contributing to overall
road safety and reducing vehicle damage.

ACKNOWLEDGMENT

We express our heartfelt appreciation to Prof. S.D.

Lokhande for his invaluable support and guidance throughout
the project, the principal of our college, and Dr. S.R. Ganorkar,
the Head of the Department of Information Technology, for
granting us permission to undertake this project and providing
us with unrestricted access to the laboratory facilities. We also
Fig 1:- The steps involved in implementing this solution wish to express our heartfelt appreciation to our project guide,
Prof. S.P. Jadhav, for dedicating time and effort to support us
The first step involves User 1 uploading an image to throughout the project. Additionally, we extend our thanks to
detect the pothole. After detecting the pothole, User 1 is the IT Department staff for their assistance. Lastly, we would
prompted with quick questions about the issue. Next, User 2 like to thank our fellow classmates for their unwavering
from the same neighborhood is asked about the challenges support and encouragement throughout the project.
they have faced due to the pothole problem, along with others.
All collected information is stored in a database and sent to REFERENCES
relevant authorities for analysis. This proposed solution has
the potential to enhance road maintenance and safety, reducing [1]. https://www.ibef.org/industry/roads-india.aspx [Last
accidents and minimizing traffic congestion. accessed: 15 August, 2020. 09:00 am.]
[2]. K. Taehyeong, and S. Ryu, “Review and analysis of
However, there are several constraints that may hinder Pothole detection meth- ods,” J. of Emerg. Trends in
the effectiveness of the Android pothole detection system. Computing and Info. Sci., vol. 5(8), pp. 603-608,2014.
These include hardware limitations, battery life, data [3]. K. Christian, and I. Brilakis, “Pothole detection in asphalt
availability, network connectivity, privacy, security, cost, and pavement images,”Adv. Engg. Info., vol. 25(3), pp. 507-
legal compliance. 515, 2011.
[4]. H. Lokeshwor, L.K. Das, and S.K. Sud, “Method for
Additionally, there are parameters that the system should automated assessment of Potholes, Cracks and Patches
include to ensure effectiveness, such as geolocation, user from road surface video clips,” Procedia-Socialand
feedback, database management, communication protocols, Behavioral Sci., vol. 104(2013), pp. 312-321, 2013
security, and user engagement. [5]. R. Karthika, and L. Parameswaran, “An automated
vision-based algorithm forout of context detection in
Overall, the success of this solution depends on the images,” Int. J. Signal and Imaging Sys. Engg., vol.11(1),
accuracy of the Deep Learning model, the quality of user- pp. 1- 8, 2018.
provided information, the responsiveness of relevant [6]. A. Krizhevsky, I. Sutskever, and G.E. Hinton, “Imagenet
authorities, and user engagement. Therefore, the system must classification with deep convolutional neural networks,”
have appropriate parameters to manage these factors and in Proc. NIPS, pp. 1097–1105, 2012.
guarantee the effectiveness of the solution. Successfully [7]. R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich
implementing this solution can improve road safety, making a feature hierarchies for accurate object detection and
significant impact on people who use the roads frequently. semantic segmentation,” in Proc. CVPR, pp.580–587,
2014.
V. CONCLUSION [8]. P. R. Pooja; Balaji Hariharan “An early warning system
for traffic and road safety hazards using collaborative
Detecting potholes on the road is crucial for ensuring crowd sourcing”.
human safety and minimizing vehicle damage. However, the [9]. Abhishek Kumar; Chakrapani; Dhruba Jyoti
pothole detection problem presents many challenges due to the Kalita; Vibhav Prakash Singh ”A Modern Pothole
varying sizes of potholes, diverse road construction materials Detection technique using Deep Learning”.
used, different traffic conditions, and changing weather [10]. Surekha Arjapure; D.R. Kalbande ” Deep Learning
scenarios. To address this issue, this research collected 1300 Model for Pothole Detection and Area Computation”

IJISRT23FEB1109 www.ijisrt.com 1484