12 VII July 2024

https://doi.org/10.22214/ijraset.2024.63499
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 12 Issue VII July 2024- Available at www.ijraset.com

Road Sense: Intelligent Road Monitoring System

Naina Nimisha1, Subrat Pandey2, Siddhant Priyadarshi3, Dr. Anusha Preetham4
Dept. of Artificial Intelligence and Machine Learning BNM Institute of Technology (Affiliated to VTU) Bangalore, India

Abstract: This project proposes a system for detecting traffic signals, lane layouts, and speed bumps in road infrastructure using
video footage through machine learning. Lane detection is performed through region of interest selection and edge detection.
Lane lines are extracted based on specific characteristics. A deep learning model is trained to detect lane boundaries and road
curvature. The system provides real-time alerts and recommendations to enhance road safety and driving experiences.
Keywords: YOLO v8, Object Detection, Lane Detection, Autonomous Vehicles, TensorFlow, Python

I. INTRODUCTION
Our advanced system combines computer vision and machine learning to enhance road. Safety and traffic management. By
accurately detecting traffic signals, lane layouts, and curves, speed bumps in real-time, our solution aims to improve driver
awareness and optimize traffic flow.
Our objective is to develop a robust real-time system for identifying and classifying various traffic signals. Through image
preprocessing and color segmentation, we isolate signal regions based on their distinct colors.
Advanced algorithms, including blob detection, distinguish signals from other objects. We extract essential features like shape, size,
position, and color to characterize the traffic signals.
Machine learning techniques, such as pattern recognition and deep learning, classify these features accurately, determining the
specific signal type.
In addition, our system includes advanced lane detection. By defining a region of interest (ROI) and utilizing Canny edge detection,
we identify lane markings with high contrast against the road surface. This enables us to extract lane lines for improved driver
assistance and lane-keeping functionality.

II. PROPOSED METHODOLOGY

A. Image Extraction and Preprocessing
The project proposes a system for detection of traffic and road signals, lane and curve layouts for roads. It involves image or frames
acquisition from video footage captured using cameras. This is followed by the image preprocessing to enhance resolution of the
image and reduce noise.

B. Feature Extraction and Segmentation

Color segmentation is done to isolate regions containing the colors of the traffic signals by setting appropriate threshold values for
hue, saturation, and value (HSV). Relevant features are then extracted from blobs to characterize the traffic signal. These features
may include shape, size, position, and color information. A machine learning model is employed to classify the extracted features
and determine the type of traffic signal. This is done using deep learning- based approaches of YOLO v8.
For lane detection, region of interest is defined to focus only on the road area. Typically, this region is a trapezoidal shape
representing the area in front of the vehicle where lanes are expected to be present. Canny edge detection is applied to detect edges
within the ROI. This helps identify lane markings that exhibit significant contrast with the road surface.
For speed bump detection, filtering the detected objects to focus on potential speed bumps based on their size, shape, and location
within the frame. We use a classifier to distinguish speed bumps from other objects in the ROI. We are determining the precise
location and dimensions of the detected speed bump.

C. Awareness Alerts
Based on the interpretations obtained from the algorithms, alerts and suggestions are provided to the driver and actions are
recommended based on the potential problems.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 224
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 12 Issue VII July 2024- Available at www.ijraset.com

III. ARCHITECTURAL DESIGN AND BLOCK DIAGRAM

This diagram signifies the main steps that are executed during the system run.

Fig – I

The below diagram shows the detailed working of the algorithms for producing successful alerts as output.

Fig – II

IV. ADVANTAGES OF THE NOVEL SOLUTION

1) Deep learning models handle complex scenarios such as lane changes, merging lanes, and intersections more effectively. They
can capture the contextual information from the surroundings, such as the presence of other vehicles and road geometry, to
make accurate predictions for different road layouts.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 225
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 12 Issue VII July 2024- Available at www.ijraset.com

2) They achieve high accuracy in lane detection from diverse labelled-datasets and powerful architectures like convolutional neural
networks (CNNs). The model can learn intricate lane patterns and generalizations that are difficult to achieve with handcrafted
rules and heuristics.
3) ML algorithms like YOLO for image classification and object detection to find traffic lights and signals help analyzing large
amounts of data in real-time for various traffic scenarios, weather conditions, and lighting variations without manual
intervention.
4) Machine learning models extract relevant features and patterns from the data, making them more robust to noise and variations
in the appearance of traffic signals and can be continuously adapt to evolving designs.
5) ML models allow integration of new features into current design to improve efficiency and can be easily deployed on different
systems.

V. MODEL TRAINING AND ACCURACY

Fig. 1. Model Training

Fig 2-Accuracy , Precision and Recall for Model

Fig 3 - Detailed graphs of the model training

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 226
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 12 Issue VII July 2024- Available at www.ijraset.com

Fig 4 - F1- Confidence Curve

Fig 5- Recall Confidence Curve

VI. WORKING OF THE SYSTEM

Fig 6 – User Interface of the System

Fig 7- Pop up window to select input file

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 227
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 12 Issue VII July 2024- Available at www.ijraset.com

Fig 8 – Stop Sign detected after processing

Fig 9 – Speed Bump detection being carried out

Fig 10 – Road Lanes detection and marking real-time

Fig 11 – Speed Limit detection and alert is sent

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 228
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 12 Issue VII July 2024- Available at www.ijraset.com

Fig 12 – Another Road Sign being detected after processing

VII. CONCLUSION
In conclusion, our road monitoring system offers significant advantages in improving road safety, traffic management, and lane
detection. Its ability to provide real-time data, enhance safety, optimize traffic flow, and inform decision- making processes makes it
a valuable tool for transportation authorities and road users alike.
The research findings have demonstrated the system's ability to accurately collect and analyze data to optimize traffic management
strategies, detect curve layouts on the road, determining lanes and provide timely warnings to drivers about potential risks and avoid
congestion on roads.
While this research is promising the benefits of this system, there may be challenges in terms of implementation, cost, and data
management. Further studies and developments to include more features and develop a more interactive interface for user while
maintaining the accuracy are essential to refine the system for releasing a prototype.

VIII. ACKNOWLEDGMENT
We would like to express our sincere gratitude to all individuals and organizations involved in this project. First to our guide Dr.
Anusha Preetham for her insightful feedback and guidance during the research process. Secondly, to Prof. Mahanthesha U for
encouraging the projects involving Digital Image Processing. Lastly, to our Artificial and Machine Learning Dept. for having a
conducive environment to promote projects and research.

REFERENCES
[1] Yanzhao Zu and Wei Qi Yan, “Traffic Sign Recognition based on Deep Learning,” Springer, March 7 2022. (references)
[2] Satish Kumar Sattia, K. Suganya Devia, Prasenjit Dhara, P. Srinivasan, A Machine Learning Approach for detecting and tracking Road Lanes, ICT Express
Vol 7 Issue 1, March 2021.
[3] Rajesh Vand Rajshekhar JS, “Vision Based Traffic Sign Detection and Recognition using Artificial Intelligence,” iJRASET, July 2022.
[4] Yueen Ma, Vincent Havyarimana, Jing Bai and Zhu Xiao, “Vision Based Lane Detection Marking Model Inference”, May 2019.
[5] M. Swathi, K.V. Suresh, “Automatic Traffic Sign Detection and Recognition in Video Sequences,” 2nd IEEE International Conference, pp. 476-480, May
2017.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 229