How to Cite:

Devi, N. B., Balaji, V. R., Beenarani, B. B., & Kala, . K. (2022). Traffic light detection and control with connected
vehicle for automated car using ML algorithm. International Journal of Health Sciences, 6(S7), 47223–
47233. https://doi.org/10.53730/ijhs.v6nS7.13115

Traffic light detection and control with

connected vehicle for automated car using ML
algorithm

N. Bharatha Devi
Assistant Professor.SG, Department of Computer Science and Engineering
Saveetha School of Engineering, Saveetha Institute of Medical and Technical
Sciences, Thandalam, Chennai

V. R Balaji
Assistant Professor, School of Electronics Engineering and Vellore Institute of
technology, Chennai Campus, Chennai

B B Beenarani
Assistant Professor.SG, Department of Computer Science and Engineering
Saveetha School of Engineering, Saveetha Institute of Medical and Technical
Sciences, Thandalam, Chennai

Kala K
Assistant Professor.SG, Department of Bioinformatics
Saveetha School of Engineering, Saveetha Institute of Medical and Technical
Sciences, Thandalam, Chennai

Abstract---The paper centers to plan a self-driving vehicle model with

the assistance of Raspberry Pi as a handling chip. The model will
perform five undertaking's that remember self-driving for track,
identifying traffic signal, rain identification, fog discovery and will help
in impediment recognition and crash evasion. The essential
information from this present reality to the vehicle is given by camera.
It utilizes a HC-SR04 following sensor module that data to the
framework through raspberry pi to our framework which is associated
with PC with a similar organization. The gathered information is then
prepared and examined and pertinent data and subtleties is made
back aware of vehicle for proper moves to be initiated. The gamble of
human blunders is in this way kept away from securely and keenly as
the vehicle can arrive at the given destination. Path Detection,
Obstacle Detection, Traffic Light Detection, Rain and Fog are
distinguished. The vehicle is halted when impediment and traffic
signal red is noticed. As a section one static is planned with Arduino
to make Bluetooth (HC-05) communication between the vehicles. This
association between two vehicles will keep away from accidents.

International Journal of Health Sciences ISSN 2550-6978 E-ISSN 2550-696X © 2022.

Manuscript submitted: 9 May 2022, Manuscript revised: 18 July 2022, Accepted for publication: 27 August 2022
47223
47224

Keywords---lane detection, raspberry pi, obstacle detection.

Introduction

Everyone, whether traveler, passenger, driver, or pedestrian would see on the

sidewalk where a different signboard provided important data. These important
road sign features help us as guides, warnings and traffic controllers. As a traffic
control tool, signals require thorough consideration, respect and proper reaction
of the driver. Road signs were discovered long ago in history. Landmarks,
direction or distance were early road signs. In the Middle Ages, multi-sided signs
at the crossroads were common, providing guidance in urban and suburban
areas. With the advent of computerized cars and their increasing pressure on the
roads, many have used common symbols and made their markings to encourage
global travel, where language comparisons may be a burden to foreigners. It is
often used to improve traffic safety through proper awareness, principles and
indicators. Many of them use symbols instead of words and are recognized and
accepted around the world.

II. Literature Survey

Implementation of lane detection algorithm for self-driving car on toll road

cipularang using Python language

Mochamad Vicky ghaniaziz , proposed “Implementation of lane detection

algorithm for self-driving car on toll road cipularang using Python language”.

In this paper they dealt with Deep Learning, Computer Vision, Fusion Sensor,
Control, until Path Planning are utilized. Every one of these are the combination
of the multitude of fields. Here in this paper we talks about the consequences of
execution of path location calculation on expressway Cipularang. With the
assistance of the camera which is put at top of the vehicle that Video will be
taken. For the picture it is utilized to deal with technique python. These technique
is combination of various strategies like variety area, line location, canny edge
identification, and Hough transform. The outcome which has stopped by playing
out the assignment shows that a few changes ought to be added like changing the
boundaries during the constantly to these calculation. Since steady boundaries
must be utilized in similar lighting conditions. The precision accomplished is
90%.

A Raspberry-Pi Prototype of Smart Transportation

Shahab Tayeb , proposed “A Raspberry-Pi Prototype of Smart Transportation”.

This paper proposed a model of a level 3 self driving vehicle utilizing Raspberry Pi,
which is utilized for identifying the close by vehicles utilizing an IR sensor. Here
the principal endeavor is to examine independent vehicle from tiny level and
spotlight on every vehicle and there correspondence with different vehicles out
and about. There are two arrangements of detached and dynamic trials has been
performed on a couple of models, which shows the interconnectivity of the created
 47225

model. What's more, numerous sensors were utilized in these copying in light of
System-on-Chip to additionally exhibit the achievability of the proposed model.

Traffic Light and Sign Detection for Autonomous Land Vehicle Using
Raspberry Pi

Priyanka D, proposed “Traffic Light and Sign Detection for Autonomous Land
Vehicle Using Raspberry Pi”. This paper, point is to carry out traffic signal and
sign recognition involving Image processing strategy for an autonomous vehicle.
In this framework errands, for example, traffic signal and sign identification for
independent vehicles has been played out The strategy which has been proposed
is utilized for sending the data in regards to when the vehicle needs to take
redirections when there will be any turns on street and begin/quit by traffic
signals. The essential data will be caught by the camera which is interacted to the
Raspberry Pi. The Images which are gathered by the camera are pre-handled with
a few picture handling procedures, for example, Hue, Saturation and Value,
variety space model strategy is utilized for distinguishing traffic signals, for the
recognition of sign Hue Saturation Value model is. The signs are which are on
street will be recognized in view of Region of Interest. The Region of Interest will
be distinguished in view of the elements like tone and state of the article in the
picture which contains traffic signs. The analysis which has led is exceptionally
exact in order of various traffic sign examples on street which has complex
foundation pictures and the outcomes achieved by playing out all errands has
decreased computational expense.

Real-Time Self-Driving Car Navigation Using Deep Neural Network

Truong-Dong Do, proposed “Real-Time Self-Driving Car Navigation Using Deep

Neural Network”.In this paper a self-driving vehicle model utilizing Deep Neural
Network on Raspberry Pi is proposed. Here in this work, a model has been which
straightforwardly maps the crude information pictures to an anticipated guiding
point as result utilizing a profound brain network calculation. Two overlay
specialized commitments are there in this work. Initial one is the Convolutional
Neural here in these model boundaries were prepared by the assistance of the
web camera which is put on top of the model, Raspberry Pi 3 Model B. The
preparation information which is gathered are the street pictures which is taken
by the camera these pictures matched with the time-synchronized directing point
and create by physically driving. Second one is the street trial of the model on
Raspberry to drive itself in the external climate, for example, oval-formed and 8-
molded with traffic sign lined track. The outcomes accomplished by playing out
these investigation are very adequacy and power of autopilot model in path
keeping task. Camera inertness is the fundamental downside. It is characterized
as the period from the time the camera sensor notices the scene to the time the
PC really peruses the computerized image information.

Deep Learning Techniques for Obstacle Detection and Avoidance in

Driverless Cars

NischalSanil, proposed “Deep Learning Techniques for Obstacle Detection and

Avoidance in Driverless Cars”. In this paper a scaled down RC which is a self-
47226

driving vehicle has been executed by utilizing Convolutional Neural Networks.

These RC vehicle has been prepared on an independent track which is 12 to 14
meters in length. The principal point of these work is to execute obstruction
location and evasion in a self-propelled vehicle. This venture utilizes a Raspberry
pi has been utilized which will control the vehicle and performing derivation
utilizing CNN, in view of its feedback. This model has performed all around well
yet when these will any sharp turns it is confronting a challenges which was
because of the impediment of the engine's turning capacities. The vehicle had
effectively recognized every one of the driving examples inside the extent of the
dataset and would require more information collection for acclimating to new
situations. These RC vehicle has got precision of 88.6% by utilizing deep learning
strategy.

Autonomous Object Detection and Tracking using Raspberry Pi

SampaJana, proposed “Autonomous Object Detection and Tracking using

Raspberry Pi”. This paper presents the execution of an ongoing vision-based way
to deal with identify and follow highlights in a climate utilizing an independent
robot which will independently tracks and distinguishes the item which will be
utilized for movement recognition of articles on a video which is gathered by robot
utilizing camera. Object recognition will be finished by finding objects in the edge
of a video which has been taken by the robot utilizing camera. Object location
instrument is utilized by each following technique in which the article initially
shows up in the video. By the better quality and savvy cameras and the rising
requirement for robotized video investigation has created a lot of revenue in object
following calculations. There are three assignments which we need to act in video
examination, initial step is location of fascinating moving items, second step is to
following of such articles from every single casing to edge, and third step is to
make investigation of article tracks to perceive their way of behaving. In this way,
the utilization of item following calculation will obtain great outcomes in the
assignments of, movement based acknowledgment, Automatic discovery.

Lane detection technique based on perspective transformation and

histogram analysis for self-driving cars

Raja Muthalagu, proposed “Lane detection technique based on perspective

transformation and histogram analysis for self-driving cars”. In this paper,
there is show of an insight calculation which will be absolutely founded on the
information which is gathered by camera. Also, it is centered around the
exhibition of path recognition strategy involving contemporary PC vision methods
for self-driving vehicles. Which presents a smaller than usual methodology in light
of edge recognition and polynomial relapse which is the methodology that is made
for recognizing just the straight path lines. Then, at that point, proposed a
superior path identification method which depends on the viewpoint changes and
histogram examination. These proposition of further developed path identification
approach is superior to the regular strategies. For exhibiting the improved
outcomes created by the proposed path recognition approach over the customary
methodology, recreation results has taken from the various conditions.
 47227

Real-time Traffic Light Detection and Recognition based on Deep RetinaNet

for Self Driving Cars

Aneesh, proposed “Real-time Traffic Light Detection and Recognition based

on Deep RetinaNet for Self Driving Cars”. In this paper for constant location
and acknowledgment of traffic signals, they proposed RetinaNet (a profound brain
network engineering) based model through the exchange learning. Here the
profound brain network RetinaNet has been utilized as model and these
framework has been executed in Keras with TensorFlow backend in Google
Colaboratory cloud stage. These RetinaNet model was prepared and tried on
Traffic Light Dataset containing traffic signal pictures of goal 1280 by 720 pixels,
which has four kind of classes. By these the model has accomplished better
precision of discovery and order than other profound learning strategies for
ongoing activity. In this proposed framework, both traffic signal identification and
its state arrangement from RGB pictures is accomplished by a solitary RetinaNet
organization. This framework can run at constant speed with further developed
precision.

Moving Object Detection and Tracking Using Convolutional Neural Networks

Shraddha Mane, proposed “Moving Object Detection and Tracking Using

Convolutional Neural Networks”. In these paper object location and following
are the primary advances which will be performed by utilizing PC vision
calculation. Here the powerful article discovery is the exceptionally large test
because of varieties in the scenes and another greatest test is to follow the item in
the impediment conditions. That is the reason in this methodology, the moving of
the items location utilizing TensorFlow article discovery API. And afterward the
area of the article which is distinguished is pass to the item following calculation.
Convolutional Neural Network based object following calculation is utilized for
these vigorous article recognition. The article can be recognized in various light
and impediment by this proposed approach. Furthermore, this methodology of
CNN has accomplished the exactness of 90.88%.

Traffic Light Detection in Autonomous Driving System

Dijana Vitas, proposed “Traffic Light Detection in Autonomous Driving Systems”.

This paper proposes a traffic signal acknowledgment framework where versatile
thresholding and profound learning are utilized for district proposition and traffic
signal restriction, individually. These proposed framework has two unmistakable
advances. The initial step is utilized for finding the locale proposition by utilizing
versatile thresholding and mass identification. Second step involves a CNN for the
exact confinement of TLs in the proposed districts. The LISA open-source dataset
is utilized alongside custom expansion strategies to expand the quantity of
accessible information tests. The exhibition of these proposed framework is
introduced as obvious and misleading positive rates acquired on the test
information. This proposed calculation gives 89.60% genuine identification rate,
while the relapse part of the model created a right area of the traffic signal in
92.67% of cases.
47228

III. Objectives

• The Aim of this project is to make the system fully supported for self-
driving vehicles, using the Raspberry Pi.

• The automotive system uses IR sensor, ultrasonic sensor, motors, rain

sensor, gas sensor.

• The purpose of a private vehicle is to determine the speed of application in

a particular route.

• Two different situations are considered: red light and green light.

• And to establishing connection between two cars arduino car and

raspberry pi car by Bluetooth for avoiding accidents.

IV. Proposed System

The main goal of this proposal is to make the project fully automated with other
new features namely, route acquisition using an IR sensor, acquisition of material
using an Ultrasonic sensor. Traffic detection through camera module and
machine learning, rain detection and fog detection finally car-to-car
communication is done using the Bluetooth module. This project provides
additional user flexibility to use fully automated user interface which enables the
user to use special side features. The stand-alone car using the Raspberry pi
contains an ultrasonic sensor, IR sensor, rain sensor, fog sensor, camera module,
16x2 LCD display, dc motor, HC-05 module. A private car using the Arduino Uno
contains a 16x2 LCD display, HC-05 module. The main feature of this model is
the Raspberry pi 3 model B + is the latest product in the Raspberry Pi 3 series,
boasting a 64-bit quad core processor running at 1.4GHz, dual-band 2.4 GHz and
5 GHz wireless LAN, Bluetooth 4.2 / BLE, fast Ethernet, and POE power with
different POE HAT. Raspberry pi 3B + uses an SDHC port or USB storage device
to boot and store files. We use the python IDLE in the programming language of
the raspberry pi model. We use Arduino software in the programming language of
the Arduino Uno pi model. The camera module is connected to the Raspberry pi 3
B + model to capture traffic signal commands to the user processed on the
blueberry board. The IR sensor is used to find a route and transmit it to the
raspberry pi module. Ultrasonic sensor is connected to the area in front of the
private car and the sensor becomes a Detect object in front of our car. The GAS
sensor is used to detect fog in a car. A rain sensor is used to detect rain.
 47229

V. Methodology

The Raspberry Pi Microcontroller is used as the backbone of this project where all
sensors including IR sensors, Ultrasonic sensors, rain sensors, gas sensors and
other devices are connected to this module. The Arduino UNO Microcontroller is a
type of controller that operates as an approaching vehicle. The project uses two
left-handed IR sensors and R Right sensors to locate the route, enabling the
autonomous vehicle to detect the route and drive properly. Ultrasonic sensor is
used for object detection. When an object is found a private vehicle will stop. A
rain sensor is used to detect rain when the person in the car will know that the
rain is coming. The gas sensor is used to detect fog in moving vehicles that helps
to detect fog easily. USB camera used to record traffic signal when the Red light is
ON the car should be STOP, when the GREEN light the car should go. A
Bluetooth device is used here to communicate a short distance, whenever a car is
found near a moving car it will know that another car is approaching near this
car. With the help of this module road accidents can be reduced. The Raspberry
Pi has a built-in Bluetooth device (HC-05) so there is no need to connect an
external Bluetooth device to this module. Arduino Uno does not have a built-in
Bluetooth device so it is connected externally. The DC-motor is used to move the
car independently with the help of which the car can move differently.

CNN is a type of neural network that feeds into AI. It is widely used for image
recognition. CNN represents data entry in the form of multidimensional arrays.
Works well with large amounts of labeled data. CNN releases each part of the
inserted image, known as the reception field. It provides the weight of each
neuron based on the critical role of the receiving field. In order to differentiate the
importance of neurons to one another .The CNN structure consists of three layers:
(1) convolutional, (2) pooling, and (3) fully connected.
47230

CNN is a deep neural network designed to analyze imagery. Recently, it was

discovered that CNN has excellent potential for successive data analysis such as
native language processing. CNN always contains two basic functions, namely
convolution and integration. The conversion function using multiple filters is able
to extract features (feature map) from the data set, where their corresponding
location information can be stored. The integration function, also called
subsampling, is used to reduce the size of the map element from the convolution
function. High integration and central integration are common integration
functions used on CNN. Thanks to CNN's interaction, the jum is a common option
for activation to transfer gradient to back-to-back training training.

Convolutional neural networks (CNNs) are widely used in many computer vision
systems such as image classification, face recognition, object detection, and more.
This chapter introduces some of CNN's most popular platforms, including LeNet,
AlexNet, VGG, GoogLeNet, and ResNet. Then, it introduced two scenarios to
launch Inception-v3, CNN's computer vision. In the first case, we create an
Inception-v3 model using Keras Applications. In the second case, we use
TensorFlow Hub to create an Inception-v3 model.

Lane Detection:

Here we used two IR sensors to find the route. Left IR and Right IR are useful for
navigation of the car so that car will move forward, backward, stop, left or right.
The IR Transmitter looks at the dark area and the IR light is visible with an IR
receiver. When the sensor receives a thread then the car controller is transferred
to move it the way forward. If the right IR sensor detects a thread, the vehicle
should move to the right. The left IR sensor receives a thread and the car should
go to the left. It works on the concept of to find a line on the track.

Here we can see in these figure forward because ir sensors has detected the lane
so that the forward message has been displayed.
 47231

Obstacle Detection:

Here we utilized HC-SR04 ultrasonic sensor is utilized for the impediment

location. It has transmitter and recipient, transmitters. Here we utilized HC-SR04
ultrasonic sensor is utilized for the deterrent recognition. It has transmitter and
beneficiary, transmitter ship off a particular repeat, when deterrent in the
distinguishing proof bearingthe beneficiary is reflected by the recipient chamber,
following dealing with through the circuit

then, at that point, the sign gives the outcome. The convincing distance range 2-
80cm, it is 3.3V-5V to work voltage. The recognizable proof extent of the sensor
can be changed by the potentiometer. The transiter conveys the message in
forward heading. Assuming that the article is before the sign, the signs get
reflected from the outer layer of the article. The location point +/ - 45. The
location point can bechanged by the potentiometer, the recognition can be
expanded by changing potentiometer in clockwise heading, and will be diminished
when changed in counter clockwiseend off a particular repeat, when block in the
ID bearing the beneficiary is reflected by the recipient chamber, right after dealing
with through the circuit then the sign gives the outcome. The convincing distance
range 2-80cm, it is 3.3V-5V to work voltage. The ID extent of the sensor can be
changed by the potentiometer. The transiter conveys the message in forward
heading. On the off chance that the article is before the sign, the signs get
reflected from the outer layer of the article. The discovery point +/ - 45. The
discovery point can be changed by the potentiometer, the discovery can be
expanded by changing potentiometer in clockwise course, and will be diminished
when changed in counter clockwise.

Here we can see in these figure that Object ahead stop has the ultrasonic sensor
has detected the object which is ahead of car and passes message to stop car.

Rain and Fog Detection:

Here we are using rain sensor for the detection of rain. And fog sensor for
detection of fog.

Here in these figure we can see raining.

47232

Here in these figure we can see fog detected.

Bluetooth module:

Here we will be using HC-05 module to make the communication between

raspberry pi car and the arduino car. So that both cars can communicate with
each other and there can be avoidance of accidents.

Here we can see car 0 detected because the here raspberry pi is car 0 it is nearing
to the arduino car. As it nears to arduino car we can see in the led display of
arduino car that car 0 has detected.

Traffic Signal Detection:

The working of the street traffic signal disclosure computation is according to the
accompanying:

It gets the edges reliably from the pi camera and sends it to the system through
Raspberry pi. It then, at that point, changes the image from RGB over totally to
grayscale, which is a norm advance in estimations using OpenCV. The attention
to the image is also different to 0.75. Then, it segments the image into different
squares and each square is then taken care of freely. It then perceives any traffic
sign models present in the image, and returns the potential gains of the bearings
for the skipping enclose (x, y, w, h) plan. At last, it shows a picture including the
traffic signal sign, passed the stopping request to the motor using 10 pins and
stop or run the vehicle. The working of the computation is in like manner sorted
out as a stream chart underneath. The dataset used given by OpenCV which
contain arranged pictures of traffic signals.
 47233

Here in these figure we can see that the green light has been detected has the
signal detected is green the car will move.

Here in these figure we can see that red light has been detected has the signal
detected is red the car will stop.

VII. Conclusion

The Self-driving vehicle was effectively ready to explore itself through the path
while identifying road sign, deterrent and path and making moves as needs be
with practically no human intercession. Raspberry pi is utilized here which
expands the exhibition and the handling speed. It utilizes both Open CV and
sensor. It tends to be seen that the arrangement proposed in this paper has better
precision and execution contrasted with different methodologies . The utilization
of sensor assists with decreasing the general expense and builds the presentation.
The independent vehicle utilizing Raspberry pi contains a ultrasonic sensor, IR
sensor, downpour sensor, haze sensor, camera module, 16x2 LCD show, dc
engine, HC-05 module. Furthermore, by creating the correspondence between two
vehicles we can keep away from the accidents.

References

[1] Mochamad Vicky Ghani Aziz; ArySetijadiPrihatmanto; HilwadiHindersah,

https://ieeexplore.ieee.org/document/832355
[2] Shahab Tayeb; Matin Pirouz; Shahram
Latifi,https://ieeexplore.ieee.org/document/8121669
[3] D Priyanka; K Dharani; C Anirudh; K Akshay; M P Sunil; S A Hariprasad,
https://ieeexplore.ieee.org/abstract/document/8365328
[4] Truong-Dong Do; Minh-Thien Duong; Quoc-Vu Dang; My-Ha Le,
https://ieeexplore.ieee.org/document/859559
[5] NischalSanil; PasumarthyAnkith Naga venkat; V. Rakesh; RishabMallapur,
https://ieeexplore.ieee.org/document/907315
[6] Sampa Jana, Shubhangi Borkar ME
Student,https://test.globalinfocloud.com/technodigisoftnew/wpcontent/uploads/201
9/07/Autonomaous-object-detection-and-tracking.pdf
[7]RajaMuthalaguaAnudeepsekharBolimera
V.Kalaichelvib,https://www.sciencedirect.com/science/article/abs/pii/S0045790620
305085
[8] A N Aneesh; Linu Shine; R Pradeep; V
Sajith,https://ieeexplore.ieee.org/document/8993293
[9] Shraddha Mane; SupriyaMangale,
https://ieeexplore.ieee.org/abstract/document/8662921
[10] Dijana Vitas; Martina Tomic;
MatkoBurul,https://ieeexplore.ieee.org/abstract/document/910941