SELF DRIVING CAR USING RASPBERRY PI

Shashank Agrawal, Vineet kumar, Vishal Vishwakarma,

Dr. Amar Kumar Dey
“Department of Electronics and Telecommunication Engineering, BIT, Durg”

ABSTRACT

Self-driving technology is becoming increasingly common and could revolutionize our

transportation system in the near future. The success of autonomous vehicles largely depends
on the development of reliable real-time detection systems for traffic control devices. This
paper proposes a self-driving car prototype that utilizes various technologies such as image
processing and machine learning, including algorithms for road lane detection, stop sign
recognition, and traffic light recognition, controlled by a Raspberry Pi controller, with an
Arduino UNO for additional processing, and an H-bridge to drive four DC motors for vehicle
automation.

Keywords- Self-driving car, Raspberry-Pi, Lane detection, Stop Sign Recognition, Traffic
Signal Recognition.

I. INTRODUCTION

Road accidents are a major problem worldwide, resulting in approximately 1.25 million
deaths annually and an average of 3287 deaths per day. Additionally, between 20-50 million
people suffer injuries and disabilities for life each year, according to the World Health
Organization (WHO). It is found that 78% of road accidents are due to driver’s fault or human
error in driving and it is interesting to address this human issue with deployment of
technology in the form of “Driverless Car”.

The idea of self-driving cars has been talked about since the 1920s, but it wasn't until 1977
that Japan developed the first semi-automated car. A significant milestone in self-driving cars
was achieved in the 1980s when a car with a speed limit of 31 kmph was created. Fast forward
to 2013, Tesla began their autonomous car project and in 2014, they introduced the Tesla S
with a semi-autopilot mode. Subsequent advancements in autopilot software were made in
later models such as the Tesla X.

The researchers have developed a new AI model called SSD-MobileNet using CNN
technology to identify and monitor real-life objects. The prototype has shown impressive
performance in detecting and tracking trained objects, boasting a 99% accuracy rate with
a confidence level of 98.2% [1]. This advancement aids in preventing accidents for robots.
Additionally, a mathematical framework has been introduced to investigate the behavior
of a two-wheeled self-balancing robot on flat and sloped surfaces [2]. A new system for
detecting lanes using Raspberry Pi and Arduino technology has been developed,
consisting of two main components: object detection and image processing. The camera
on the prototype captures images, which are then analyzed to identify lane markings. The
system also includes an obstacle detection feature that uses data from the images to
generate commands. While this system has some limitations in accurately detecting lanes
[3], it is able to recognize and respond to traffic signs and lights in real-time [4].

The paper is structured as follows: In Section II, the design of the self-driving car is explained
using its circuit diagram. Section III discusses the subsystems of the self-driving car and their
functions. Section IV covers the results and discussion, while Section V offers conclusions and
suggestions for future work.
II. ARCHITECTURE OF SELF-DRIVING CAR

The circuit diagram of the self-driving car is shown in the Fig. 1.

Fig. 1: Circuit Diagram of Self Driving Car

The components used in this project are enlisted below:

1.Raspberry Pi: The Raspberry Pi serves as the central control unit for the project, processing
sensor data and running algorithms for autonomous navigation. It includes a CPU for
computing, RAM for temporary storage, and an SD card for software and resources. GPIO pins
allow communication with peripherals such as sensors and motors, while the camera module
captures real-time images for lane detection and traffic analysis. USB ports provide
connectivity with external devices, and networking capabilities allow communication with
other systems. The Raspberry Pi, with its operating system managing resources, serves as a
flexible platform for developing the self-driving car system.

2. Arduino UNO: The Arduino Uno is the microcontroller for the project, receiving commands
from the Raspberry Pi to manage the motors of the vehicle. It has digital and analog I/O pins
for connecting to sensors and motor drivers, PWM pins for controlling motor speed, and UART
communication for receiving navigation instructions. The Arduino Uno maintains stable
operation with voltage regulation and has a reset button for troubleshooting. It effectively
controls the vehicle's movements based on input from the Raspberry Pi for autonomous
navigation.

3. L298N Motor DriverThe L298N motor driver is a key part of the self-driving car project,
allowing for two-way control of DC motors. Its dual H-bridge setup, ability to handle high
currents, PWM speed control, and diode protection make it perfect for efficient and
dependable motor control during autonomous navigation. It works seamlessly with the
Arduino Uno, allowing for precise control of the vehicle's movement to ensure smooth
operation.

In order to control all four motors at the same time using the L298N motor driver, a paired
connection setup is used. This involves grouping the motors into pairs and connecting each
pair to a single L298N motor driver module. By using a cross-connection arrangement, where
one motor from each pair is connected to each output of the motor driver, all four motors
can be controlled simultaneously. This setup ensures that the motors operate in sync,
allowing the vehicle to move in a coordinated manner during autonomous navigation.

4. IC74HC164N SIPO: The IC74HC164N, also known as a Serial-In-Parallel-Out (SIPO)

integrated circuit, plays a key role in converting serial input data into parallel output signals.
In our self-driving car project, this IC is crucial for controlling the vehicle's lighting system.
When the car needs to stop, signals from the Arduino Uno are processed in series by the
IC74HC164N. This results in all LEDs turning on at once, giving the appearance of a brake light.
Likewise, during turns, the IC lights up LEDs on the correct side one by one, serving as a
sequential turn signal. This setup ensures that the vehicle's lighting system works together
smoothly, improving safety and efficiency when the vehicle is navigating autonomously.
III. SUB-SYSTEMS OF SELF-DRIVING CAR

A. Lane Detection System (LDS):

To achieve the main objective of lane detection, a self-driving car should be able to
detect the tracks or lanes in the road. The Lane Detection System consists of a camera
module mounted on top of the self-driving car which is connected to the Raspberry Pi
controller to detect the position of the car relative to the white lines or the tracks on
each end of the road. The process flow of Lane detection system is shown in Fig. 2.

In the proposed self-driving car, the lanes at each end of the road are designed which
the self-driving car will follow by remaining at the centre of the road. When the
camera is off, no commands will be sent to the Arduino which is acting as a slave for
controlling the motors, the Arduino module has been programmed in such a way that
according to the information or the data which the Raspberry Pi controller or the
master of the system provides it will drive the motor accordingly. When the user starts
the camera, the system starts capturing the images and creates a region of interest
(ROI) in the captured images for the detection of the lanes, after that all the images
captured are converted from RGB to GRAY scale image and then the pixels of the
image are divided (for example :- a 100 pixels image is divided into 100 individual pixel
i.e. it will have 1 image for every individual) such that we can detect the white lanes
in the road as the pixels for the white lanes will be high and the pixels for rest of the
image will be low, these whole image processing is done internally in the Raspberry Pi
controller, now from these detected lanes the system creates an imaginary line which
remains at the centre of the road (Lane Centre) and will be further used to keep the
car at the centre of the lane.

After the lane detection process is completed using image processing the next step
comes to keep the car at the centre of the lane, which is done by creating an imaginary
line which will be interpreted by the system as the car, the system will compare this
line with the lane centre to ensure the car remains in the centre of the road, prompting
the car to move forward. Whenever the car deviates from the centre, either to the
right or left, the Raspberry Pi controller sends the data accordingly to the Arduino. The
Arduino then controls the motor power based on this data to keep the car centered.
As a result, while moving, the car is always in a self-correcting mode to maintain its
position in the centre of the road.

Fig. 2: Process Flow of Lane Detection System

 Fig. 3: Implementation of LDS in Raspberry Pi controller

B. Traffic-Light and Stop Sign Detection System (TLSSDS):

A traffic-light and stop sign detection system is a vital technology in development of
autonomous driving car and systems like advanced driver assistance systems (ADAS).
In the proposed self-driving car, we will be using machine learning (ML) technology to
accurately identify stop signs in various driving environments. The process flow of
TLSSDS with LDS is shown in Fig. 4.
A traffic-light and stop sign detection system using ML works by processing and
analyzing data of the images captured by the camera mounted on the car, the system
utilizes a XML file, which stores trained data from positive and negative samples of
stop sign and traffic light.
Positive samples: These are the instances that the model learn to identify as the target
or the instance which is to be detected. For example, in a model designed to identify
spam emails, the spam emails are the positive samples.
Negative samples: These are the instances that the model learn to identify as not
belonging to the target or the instance which are not to be detected. For example, in
a model designed to identify spam emails, the legit emails are the negative samples.

The detection of stop sign and traffic light takes place in several steps: -
1. Data Collection: A wide range of images featuring stop signs and traffic light which
is to be detected (red light) are captured from different angles, lighting conditions,
and backgrounds to ensure that system is exposed to different real-time scenarios,
these images will be considered as the positive samples.

Similarly, the negative samples are also captured in different angles, lighting
conditions. In our proposed work the negative samples are the designed road
which the car is intended to follow by lane detection.

2. Preprocessing: Once collected, the images are pre-processed. This involve resizing,
enhancing contrast, cropping and to convert the image from RGB to GRAY-SCALE
image to highlight the stop signs and the traffic lights. This step makes the data
more uniform and easier for the system to analyze and utilize for training.

3. Model Training: The processed images are then used to train a detection model,
The training helps the model to learn what stop signs and traffic light looks like
from different angles and in different lighting conditions.

At last, the trained model is saved into a file in XML format, which will be further
used in our project to help the car detect and respond to the models in real time,
as it drives.

4. Model Testing: To ensure that the trained data is reliable, the trained model
undergoes a test with new images it hasn’t seen before during the training.

This step is crucial as it helps to ensure that the model can accurately identify the
models in different real-world conditions.

5. Deployment: The trained model is then incorporated within our system and
integrated to work simultaneously with the lane detection system. While the
system is busy detecting the lanes by processing the live feed from the camera
module, it also ensures that the stop sign and traffic light is detected and provide
necessary data to be used further in the design.
 This integration n of the stop sign and traffic light detection and the lane detection
ensures that our system is capable of handling multiple tasks at once, like staying
in the centre of the lane while also detecting the required models.

6. Real-Time Detection: As the car drives, performing multiple tasks of lane

detection, stop sign detection and traffic light detection at once, it also performs
another task which is the most important part of this whole system, i.e. calculating
the distance between the car and the models which is to be detected, the system
uses the linear equation y = mx + c to calculate the distance:
 x represents the change in x-axis, calculated as P2.x – P1.x, where P2
indicates the car’s position and P1 indicates the position of the model to
be detected.
 m represents the rate at which y changes with respect to x.
 y is distance to be calculated between car and the model.
To calculate the distance between them, we first manually calculate the distance
by placing the models at two different distances (15 cm and 30 cm) and recording
the corresponding value of x in pixels format which is displayed at the user’s
display, by using these measurements, we can determine the coefficients m and c
in the equation.

Once the system can calculate the distance between the car and the model, any
model which is detected within a 45 cm range gives a response. Based on this
response, the Raspberry Pi controller then provides the information to the
Arduino, which then execute a programmed task based on the data received from
the Raspberry Pi controller. If a stop sign is detected, the car stops for few seconds
and resume the normal operation (Fig. 5). Similarly, if a red light is detected at a
traffic light, the car stops there and wait until the signal goes green before
proceeding (Fig. 6 & Fig. 7). This multi-tasking capability of our self-driving car
ensures that the car responds appropriately to the traffic rules, while enhancing
safety.
Fig. 4: Process flow of Stop Sign and Traffic Light Detection System with Lane Detection
System
 Fig. 5: Stop Sign Detection with Lane Detection System in Raspberry Pi controller

Fig. 6: Traffic Light Detection (Red Light) with Lane Detection System in Raspberry Pi
controller
Fig. 7: No detection of Green Light and the normal operation continues
IV. RESULT & DISCUSSION

A prototype of self-driving car is developed in this project and is shown in Fig. 8. Our self-
driving car uses advanced image processing techniques to accurately detect the lanes and
ensures that it can drive safely under various conditions, the technique is highly effective for
lane detection but in rainy or foggy weather the efficiency of the system reduces. In these
cases, incorporating new sensors like RADAR or LIDAR could enhance the detection accuracy
of the system. This system also features detection of stop sign and traffic light detection, the
system is trained with wide range of images from different angles and different conditions.
Thus, the system provides a high level of accuracy in detecting stop sign and traffic lights,
ensuring effectiveness even when these signs are partially visible or at a distance. Integrating
these techniques in our system enables the car to make real-time decisions essential for safe
driving.

It is demonstrated that the system accurately follows the lane and while moving forward,
ensures the detection of the stop sign and traffic-light detection, the response of the system
to detection is as follows: for stop sign, the car stops it’s movement for few seconds and after
that resumes the normal operation, for red signal in the traffic light, the car stops its
movement until the light is red, when the red signal turns green the car then starts moving
displaying a very good accuracy and performance.
Fig. 8: Prototype of Self-Driving Car
V. CONCLUSIONS & FUTURE SCOPE

A. CONCLUSIONS

A low-cost prototype of a Self-Driving Car model is designed developed and all

functionalities are successfully demonstrated. The car is able to follow lane efficiently using
the programmed algorithms and the traffic colours are detected and decisions are made by
the car using image processing techniques to follow real-time traffic rules. Car is able to
respond to the given instructions precisely and is able to detect stop signs and respond
accordingly.

B. FUTURE SCOPE

 The lane detection algorithms can be further optimized to increase the efficiency of
the system.
 Features such as GPS can be added so that the live location of the car can be
accessed in real time for safety purposes.
 Accident alert system can be added to the car so that it contacts the emergency
services and sends the last location if the car meets with an accident.

REFERENCES
[1] G. Chandan, A. Jain, H. Jain and Mohana, "Real Time Object Detection and Tracking Using
……Deep Learning and OpenCV," 2018 International Conference on Inventive Research in
……Computing Applications (ICIRCA), Coimbatore, India, 2018, pp. 1305-1308.

[2] Pannaga R.M. and B.P. Harish, “Modelling and implementation of twowheel self-
…..balance robot,” International Journal of Electrical, Electronics and Computer Science
…..Engineering, Vol. 4, Issue 6, pp. 33- 40, December 2017.

[3] Tiple, A. Hemant, T. Hemant, Gurav, and S. Hanumant, “Prototype of Autonomous Car
……Using Raspberry Pi,” International Journal of Engineering Research in Electronics and
……Communication Engineering (IJERECE), vol. 5, 2018, pp. 39-43.

[4] T. Moura, A. Valente, A. Sousa and V. Filipe, "Traffic Sign Recognition for Autonomous
……Driving Robot," 2014 IEEE International Conference on Autonomous Robot Systems and
……Competitions (ICARSC), Espinho, Portugal, 2014, pp. 303-308.