REAL-TIME ROAD DAMAGE
DETECTION
AND GEOSPATIAL MAPPING USING
YOLOv8
Group members:
1.Asavari Bhelawe (5021106) IT Department Group 18
2.Vedika Pagar (5021138) Internal Guide: Dr. Archana Shirke
3.Aman Singh (5021159) External Guide : Dr. Shashikant Dugad
Abstract
Modern technology has transformed urban road monitoring using
video sources like smartphones, car cameras, and surveillance
systems. This project presents a scalable deep learning system for
real-time road damage detection, classification, and geospatial
mapping. Utilizing YOLOv8 model is trained on annotated
datasets and enhanced with data augmentation to adapt to
varying conditions. A user-friendly interface offers real-time
detection, mapping, and automated report generation to assist
authorities in efficient infrastructure management. This system
significantly improves road maintenance efficiency, making
damage detection 70-80% more effective and 95% faster than
traditional methods.
Introduction
Road damage, including cracks, potholes, and surface deterioration, significantly impacts safety and
transportation efficiency. Timely detection and repair are crucial to prevent accidents and reduce
infrastructure costs. Traditional inspection methods are time-consuming, subjective, and prone to errors.
Recent advancements in deep learning and computer vision offer scalable solutions for automated road
damage detection. This project uses YOLOv8, a deep learning model, to achieve real-time damage
detection with high accuracy. The integration of geospatial mapping helps visualize and prioritize repairs,
making road maintenance more efficient and cost effective.
Problem Statement
Traditional road damage detection methods rely on manual inspections and costly technologies like laser
scanning, which are slow, labor-intensive, and inconsistent. These approaches struggle with scaling across
large road network sand lack real-time monitoring capabilities. Additionally, image quality variations
due to lighting and weather conditions hinder accurate damage identification. Therefore, there is a need
for an automated, real-time, and scalable solution that can accurately detect, classify, and map road
damage, enabling faster and more effective maintenance.
YOLOv8 Model Architecture
🔹 Model Components:
Backbone: Extracts key features from input images.
Neck: Enhances object detection accuracy.
Prediction Head: Generates bounding boxes and classification scores. Outputs the detected
road damage with confidence levels.
Proposed Solution
🔹 Objective:
To develop a real-time, scalable, and accurate road damage detection system using YOLOv8. Integrate geospatial
mapping to visualize damage locations for better road maintenance management.
🔹 Key Components:
Road Damage Detection
Geospatial Mapping
User Interface.
Dataset
The images in the RDD2020 dataset are in JPEG format.
The images have a height of 600 pixels and a width of 600 pixels.
The dataset includes images from the following countries:
Japan: 9,053 images
India: 8,401 images
Czech: 3,538 images
The dataset is annotated with the following classes:
D00: Longitudinal Cracks
D10: Transverse Cracks
D20: Alligator Cracks
D40: Potholes
The annotations are provided in the PASCAL VOC format.
05
Implementation Details
07
Results & Detection Examples
The proposed model effectively identifies and classifies
various types of road damage with high confidence. The
system offers greater efficiency compared to manual
inspection methods and reduces the time required for road
monitoring. Its real-time performance ensures faster
decision-making, enabling authorities to swiftly prioritize
and address road maintenance needs.
Results & Detection Examples
The outcomes of the proposed system, accurately identifying various
types of road damage with bounding boxes and labels:
Longitudinal crack (Class: D00) highlighted with a red box.
Potholes (Class: D40) detected with green boxes.
Transverse crack (Class: D01) marked with a yellow box.
Combination of transverse and alligator cracks (Class: D10, D20)
accurately detected and labeled.
The bounding boxes and labels indicate the damage class and showcase
the model's detection accuracy and versatility across different road
conditions.
Application Interface
The application interface has two main login, one
for the common man and other for the admin.
User Interface of common man has four main
options:
Live Detection
Manual Report
Optimized Route
Road Damage Map
09
Reporting Feature
Reporting can be done from the user
side
User can add Images as well as
Description , Location is
automatically feteched from user
device location
As soon as any emergency report are
made admin get this kind of SMS
notification on registered admin
mobile number
Implemented using Twilio
Working on Push notification via
Firebase Cloud Messaging
Application Interface
The Admin gets the Automated
report generated for all the
reports that are made using the
application, they can also
prioritize which damage needs
immediate attention and
accordingly allot the labor for
repair.
Future Scope
1. Enhanced Accuracy and Resilience: Future research should focus on improving the accuracy of road damage
identification algorithms.Efforts are needed to make models more resilient to varying road surfaces and diverse
environmental conditions.
2. Exploration of Advanced Deep Learning Models: Implementation of more sophisticated models such as instance
segmentation and semantic segmentation could provide deeper insights into the characteristics of road damage.
3. Integration of Comprehensive Road Condition Databases: Utilizing large-scale, annotated datasets of road
conditions can help in training more robust models.This integration can support better-informed and timely
maintenance decisions.
4. Field Testing and Collaboration with Authorities:Real-world testing is crucial to validate system performance and
reliability.Continuous collaboration with road maintenance departments will ensure that the developed solutions
are practical and aligned with field requirements.
5. Development of Intelligent Infrastructure Management Systems:The long-term objective is to create smart, data-
driven infrastructure management solutions that support efficient and predictive road maintenance.
Conclusion
The proposed system offers a real-time, scalable, and accurate solution for road damage detection and geospatial
mapping using YOLOv8. The integration of geospatial mapping enables authorities to visualize damage locations,
helping in efficient infrastructure management and repair prioritization. The system demonstrates high improvement
in efficiency and reduces the time required for road inspections by comparing it to traditional methods. This
solution has the potential to significantly enhance road maintenance processes, optimize resource allocation, and
contribute to safer and better-managed road networks.
Appendix
[1] Road Damage Detection Using Deep Neural Networks with Images Captured Through a Smartphone Hiroya
Maeda∗ , Yoshihide Sekimoto, Toshikazu Seto, Takehiro Kashiyama, Hiroshi Omata University of Tokyo, 4-6-1
Komaba, Tokyo, Japan
[2] Kulambayev, Bakhytzhan & Nurlybek, Magzat & Astaubayeva, Gulnar & Tleuberdiyeva, Gulnara & Zholdasbayev,
Serik & Tolep, Abdimukhan. (2023). Real-Time Road Surface Damage Detection Framework based on Mask R-CNN
Model. International Journal of Advanced Computer Science and Applications. 14. 10.14569/IJACSA.2023.0140979.
[3] Wan, F., Sun, C., He, H. et al. YOLO-LRDD: a lightweight method for road damage detection based on improved
YOLOv5s. EURASIP J. Adv. Signal Process. 2022, 98 (2022). https://doi.org/10.1186/s13634-022-00931-x
[4] H. Maeda, Y. Sekimoto, T. Seto, T. Kashiyama, and H. Omata, “Road Damage Detection and Classification Using
Deep Neural Networks with Smartphone Images,” Comput. Aided Civ. Infrastructure Eng. (CACAIE), vol. 33, no. 12,
pp. 1127–1141,
2023
[5] J. Redmon and A. Farhadi, “Yolo9000: Better, faster, stronger,” in The IEEE Conference on Computer Vision and
Pattern Recognition (CVPR),2022, pp. 6517–6525.
[6] Alexey Bochkovskiy, Chien-Yao Wang and Hong-Yuan Mark Liao,“YOLOv4: Optimal Speed and Accuracy of
Object Detection,” 2020. [Online]. Available: arXiv:2004.10934.
[7] Deeksha Arya, Hiroya Maeda, Sanjay Kumar Ghosh, Durga Toshniwal, Yoshihide Sekimoto, RDD2020: An
annotated image dataset for automatic road damage detection using deep learn ing, Data in Brief, Volume 36, 2021,
107133, ISSN 2352-3409, https://doi.org/10.1016/j.dib.2021.107133.
ank Yo
h
T
