Volume 9, Issue 2, February – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24FEB1512

Machine Learning Based on Road Condition

Identification System for Self-Driving Cars
Ligitha Sakthymayuran 1; Disne Kajanath2
Trincomalee Campus, Eastern University of Sri Lanka

Abstract:- Modern self-driving cars heavily rely on visual dependability. It must respond quickly and with great precision.
inputs to make decisions and it contains resolving Modern, cutting-edge technology is still unable to fully
significant computer vision issues. The development of deep automate safe operation. Therefore, it is still necessary to
learning has opened up a number of opportunities to confirm if the object segmentation performance exceeds human
enhance those computer vision issues and hence be able to capabilities [4].
enhance performance in autonomous driving applications.
The primary function of vision-guided systems is object With new technologies, the industry must expand day by
segmentation to comprehend the surroundings. This study day, yet the issue is still unresolved in deep learning research.
uses deep learning techniques to create an effective model The goal of this proposal is to use deep learning to identify new
of the best path to follow an item on a self-driving vehicle. techniques to improve performance. The era of safe and
And helping with improved decision-making to locate the comfortable transportation in the future of autonomous vehicles
least expensive routes during navigation. is ambitious. Throughout this research, we hope to enhance
performance through a more accurate road segmentation task.
I. INTRODUCTION Testing on datasets other than prerecorded datasets, improving
human-labeled data, and other issues are part of current
In recent years, modern transportation has advanced to research concerns. In this proposal, the literature review
self-driving cars as intelligent robots thanks to the rapid includes background information about self-driving
development of Artificial intelligence. Even though there are automobiles. Next, we suggested a research action plan in the
many successful autonomous vehicles on the market today [1], methodology section along with the research's anticipated
[2], [3] studies are still being conducted to improve their contributions.
capabilities [4]. The development of machine learning has been
largely responsible for the success of self-driving cars [5], [6]. II. LITERATURE REVIEW
With the use of machine learning, many functions have been
improved, including vision-based guiding to move about the The concept of self-driving automobiles was initially
surroundings precisely [6], identifying traffic signals [7], proposed in the early 1990s, and the radio-controlled car was
proposing the least expensive route, and more. Since self- the first historical example to be presented[8]. The technology
driving cars have emerged as the newest trend in transportation, has been created for today's requirements with the help of
there are still plenty of things that may be improved. The several enterprises, universities, and research facilities. The
purpose of autonomous vehicles in the future is to improve Defense Advanced Research Projects Agency (DARPA)
pedestrian and passenger road safety, safe navigation and seedling project known as DARPA, an Autonomous vehicle,
parking, and travel time reduction by finding the fastest routes. brought an artificial intelligence-related autonomous vehicle to
It also aims to regulate traffic flow generally and minimize the developmental domains[10]. Since then, several
environmental pollutants indefinitely [8], [9]. autonomous cars have been launched as the primary study field
of artificial intelligence by Google, Tesla, and other significant
Countries and government organizations keep manufacturers, researchers, and technology businesses[1], [3],
anticipations for higher level accuracies in automobiles in the [11].
future since numerous challenges come along as a result of their
growth. However, autonomous vehicles have benefits like high As in the technological aspects, vision-based navigation
reliability, high speed, lower government spending on traffic enables a lot of features for self-driving. It includes several
enforcement, less need for vehicle insurance, fewer redundant specific challenges compared to other automated robot types
passengers, etc. [4], [8], [9], they also have drawbacks like the such as high accurate decision making with correct
need to establish a legal framework and potential misuse by identification of surroundings. In the environment,
criminals and terrorists. Since the car manufacturer is understanding includes object recognition, object localization,
responsible for the legal frame, the accuracy of autonomous distance measurement, object tracking, etc. Further, the objects
vehicles is essential. The precision of the system's are in wide ranges and are identified for different decision
understanding of the environment has a major impact on its makings on vehicle navigation, Pedestrian, moving vehicles,

IJISRT24FEB1512 www.ijisrt.com 1923

Volume 9, Issue 2, February – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24FEB1512

obstacles, and traffic signals. Road identification is some

crucial detection and the object recognition system has a high
weight on the accuracy of the overall system. In today’s
autonomous vehicles, the perceptions through large sensory
systems such as LiDAR and radar sensors with structured light,
laser scanners, ultrasound, or time-of-flight, provided the vision
to the robotic system[12]. Based on those sensors many data
sets are collected[13], [14] and used for developing models for
environment understanding[15]. Using many approaches,
object segmentation has been improved. 2D object detection,
3D object detection, and 2D and 3D Image segmentation have
been widely used for vision-based navigation systems. And
many types of image types such as Stereo images, 3d Point Fig 1: Basic System Model of the Project
clouds, and fish eye view images are used for image
segmentation. And further to obtain high accuracy in real-time For this project, we get only the number of vehicles and
with a high frame rate as well as with low cost, now the average speed of vehicles during the 30 minutes time period.
approaches are toward RGB-based image The figure shows how servers keep the data according to days
segmentation. Hence, this research will be aimed at RGB- and times.
based image segmentation for reducing gaps and inaccuracies
in existing systems. Table 1: Sample Data Set
In the same way, the other decision-making tasks have
achieved high accuracy through machine learning and deep
learning. Path planning with minimum cost, and reduce risk in
navigation necessary decision making and also needs to be real-
time. Hence, deep learning techniques such as CNN, RL based
research are carried on to improve the systems. It has been
identified that the accuracy improvement in decision-making
systems still heavily depends on the environment
understanding; hence, this research is focused on
improving road condition understanding with image
segmentation. Also, to enhance the overall decision-making in This project gives a new model to predict the time that can
identifying traffic and improve the cost-effective path selection be taken to pass a specific road. This will improve the
in the navigation. productivity of autonomous vehicle systems that reduce the
time consuming of the travel from one place to another place.
III. METHODOLOGY One of the weaknesses of this project is time consuming to
gather and process the data.
In this project, autonomous cars get data from wireless
sensor nodes that are situated in every starting point of the Throughout this project, we hope to enhance performance
roads. Mainly they provide the current situation of the road with through a more accurate selecting road segmentation. Testing
the number of vehicles in between the starting point to end on datasets other than pre-recorded datasets, improving sensor-
point. These sensor nodes collect these data from labeled data.
communication with each sensor node that number of vehicles
came into the road and vehicles went out from the road in a In this project, we gather the data from three different
specific time period which is 30 minutes. roads that start from the same position and end at the same
position. As the below figure, number of vehicles and average
These sensor nodes send their data every 30 minutes to a speed of through the road in every 30 minutes of day up to 27
cloud-based server to store the data. The server stores the data weeks. They are mentioned as road 01, road 02 and road 03.
according to the day and time which is in 30 minutes. This
server database keeps data with every day of the week within a
time period up to recent 27 weeks.

IJISRT24FEB1512 www.ijisrt.com 1924

Volume 9, Issue 2, February – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24FEB1512

Above figure shows the basic diagram of three roads. We

assume that these three roads’ distances are 25km, 35 km and
26 km respectively. On these roads, vehicles can travel 90 km/h
maximum speed and 10 km/h minimum speed.

When a vehicle comes to this junction, the vehicle gets the

current data from the sensor nodes and processes this data with
past data from the cloud server. the flow chart of this process is
as flows;

Fig 2: Assumed Three Roads for Project

Fig 3: Flow Chart of the Project

When data is provided from the sensor nodes to the IV. RESULTS
vehicle, the vehicle finds the data from the cloud server
according to the day and time period. Then the machine As the dry run, we provide the data to the system as
learning technique starts to predict the average speed through follows;
the road according to the vehicles passing over these roads at
the current time. Using the average speeds of the roads, the  Number of vehicles in each road: 10 (We assumed all the
system will calculate the time taken to travel through the roads. road have same number of vehicles)
 Day: Monday
When processing the data according to the current  Time Period: 10.30 am to 11.00 am
situation, the system checks the correlation of data. The system
will reduce the cost of the data using gradient descent The results for the above inputs were shown in the figure
technique. below. From these results, the system can predict the road that
vehicles need to travel within minimum time.

IJISRT24FEB1512 www.ijisrt.com 1925

Volume 9, Issue 2, February – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24FEB1512

[4]. “A new standard for accident simulations for self-

driving vehicles: Can we use Waymo’s results from
accident simulations? | SpringerLink.”
https://link.springer.com/article/10.1007/s00146-022-
01495-y (accessed Jul. 15, 2022).
[5]. M. Bojarski et al., “End to End Learning for Self-
Driving Cars.” arXiv, Apr. 25, 2016. doi:
10.48550/arXiv.1604.07316.
[6]. I. Krešo, D. Čaušević, J. Krapac, and S. Šegvić,
“Convolutional Scale Invariance for Semantic
Segmentation,” in Pattern Recognition, Cham, 2016, pp.
64–75. doi: 10.1007/978-3-319-45886-1_6.
[7]. F. Bohm and K. Häger, Introduction of Autonomous
Vehicles in the Swedish Traffic System : Effects and
Changes Due to the New Self-Driving Car Technology.
2015. Accessed: Jul. 15, 2022. [Online]. Available:
http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-
253996
[8]. K. Bimbraw, “Autonomous Cars: Past, Present and
Future - A Review of the Developments in the Last
Century, the Present Scenario and the Expected Future
Fig 4: Results of Autonomous Vehicle Technology,” ICINCO 2015 -
12th Int. Conf. Inform. Control Autom. Robot. Proc.,
V. CONCLUSION vol. 1, pp. 191–198, Jan. 2015, doi:
10.5220/0005540501910198.
Throughout this project, a depth study will be carried out [9]. E. R. Teoh and D. G. Kidd, “Rage against the machine?
on the current state of the art in surrounding identification. To Google’s self-driving cars versus human drivers,” J.
conclude this project, we have collected the data from sensor Safety Res., vol. 63, pp. 57–60, Dec. 2017, doi:
notes and pre-processed the data by providing day and time 10.1016/j.jsr.2017.08.008.
from this. When the vehicle selects the time period for traveling [10]. R. Behringer et al., “The DARPA grand challenge -
it will help to predict the average speed use of machine learning development of an autonomous vehicle,” in IEEE
techniques which leads to calculating the least cost path and Intelligent Vehicles Symposium, 2004, Jun. 2004, pp.
minimum time. 226–231. doi: 10.1109/IVS.2004.1336386.
[11]. M. Paul and B. Thomas, “A Review on Google Self
The future work on this project can be extended to reduce Driving Car Technology,” Int. J. Eng. Res. Technol.,
the accident rates by identifying the minimal traffic path and vol. 4, no. 6, Art. no. 6, May 2018, doi: 10.17577/IJERT
vehicles are able to choose the quality road for their travel as CONV4 06006.
well as the fuel consumption and energy consumption of the [12]. H. A. Ignatious, H.-E.- Sayed, and M. Khan, “An
vehicle overview of sensors in Autonomous Vehicles,”
Procedia Comput. Sci., vol. 198, pp. 736–741, Jan.
REFERENCES 2022, doi: 10.1016/j.procs.2021.12.315.
[13]. “The KITTI Vision Benchmark Suite.”
[1]. “Trust in autonomous vehicles: The case of Tesla http://www.cvlibs.net/datasets/kitti/ (accessed Jul. 15,
Autopilot and Summon | IEEE Conference Publication | 2022).
IEEE Xplore.”
[14]. “TORCS Dataset - AI牛丝.”
https://ieeexplore.ieee.org/document/8122757
https://www.ai2news.com/dataset/torcs/ (accessed Jul.
(accessed Jul. 15, 2022).
15, 2022).
[2]. L. Gomes, “When will Google’s self-driving car really
[15]. H. A. Elkholy, A. T. Azar, A. S. Shahin, O. I.
be ready? It depends on where you live and what you
Elsharkawy, and H. H. Ammar, “Path Planning of a Self
mean by ‘ready’ [News],” IEEE Spectr., vol. 53, no. 5,
Driving Vehicle Using Artificial Intelligence
pp. 13–14, May 2016, doi:
Techniques and Machine Vision,” in Proceedings of the
10.1109/MSPEC.2016.7459105.
International Conference on Artificial Intelligence and
[3]. D. L. Rosenband, “Inside Waymo’s self-driving car: My
Computer Vision (AICV2020), Cham, 2020, pp. 532–
favorite transistors,” in 2017 Symposium on VLSI
542. doi: 10.1007/978-3-030-44289-7_50.
Circuits, Jun. 2017, pp. C20–C22. doi:
10.23919/VLSIC.2017.8008500.

IJISRT24FEB1512 www.ijisrt.com 1926