E-BIKE SPEED CONTROL SYSTEM

A Project Report Phase – II

Submitted in partial fulfilment of the requirements for the award of Bachelor

of Engineering degree in Electronics and Communication

Engineering by

B. RAMAKRISHNA REDDY (39130082)

CH. RAMU NAIDU (39130095)

DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING

SCHOOL OF ELECTRICAL AND ELECTRONICS

SATHYABAMA
INSTITUTE OF SCIENCE AND TECHNOLOGY
(DEEMED TO BE UNIVERSITY)
Accredited with Grade “A” by NAAC
JEPPIAR NAGAR, RAJIV GANDHI SALAI, CHENNAI - 600 119

APRIL – 2023

1
 SATHYABAMA
INSTITUTE OF SCIENCE AND TECHNOLOGY
(DEEMED TO BE UNIVERSITY)
Accredited with “A” grade by NAAC
Jeppiaar Nagar, Rajiv Gandhi Salai, Chennai – 600 119
www.sathyabama.ac.in

DEPARTMENT OF ELECTRONICS AND COMMUNICATION

ENGINEERING

BONAFIDE CERTIFICATE
This is to certify that this Project Report is the Bonafide work of
B.RAMAKRISHNA REDDY (39130082), CH. RAMU NAIDU (39130095) who
carried out the project entitled “E-BIKE SPEED CONTROL SYSTEM” under the
supervision of Dr.S. Barani. M.E., Ph.D. from November 2022 to April 2023.

Internal Guide

Dr. S. Barani M . E . , Ph.D.

Head of the Department

Dr. T. RAVI, M.E., Ph.D.,

Submitted for Viva voce Examination held on

Internal Examiner External Examiner

2
 DECLARATION
We B. RAMAKRISHNA REDDY (39130082) and CH. RAMU NAIDU (39130095)
hereby declare that the Project Report entitled “E-BIKE SPEED CONTROL
SYSTEM” done by us is submitted in partial fulfilment of the requirements for the
award of Bachelor of Engineering degree in ELECTRONICS AND
COMMUNICATION ENGINEERING.

DATE: 1.
PLACE: CHENNAI 2.

SIGNATURE OF THE CANDIDATES

3
 ACKNOWLEDGEMENT

We are pleased to acknowledge our sincere thanks to Board of Management of

SATHYABAMA for their kind encouragement in doing this project and for
completing it successfully. We are grateful to them.

We covey our thanks to Dr. N.M. NANDHITHA, M.E., Ph.D., Dean, School of
Electronics and Communication Engineering and Dr. T. RAVI, M.E., Ph.D., Head
of the Department, Dept. of Electronics and Communication Engineering for
providing us necessary support and details at the right time during the progressive
reviews.

We would like to express our sincere and deep sense of gratitude to our Project
guide Dr.S. Barani,M.E.,Ph.D for her valuable guidance, suggestions and constant
encouragement paved way for the successful completion of our project work.

we wish to express our thanks to all Teaching and Non-Teaching staff members of
the Department of Electronics and Communication Engineering who were
helpful in many ways for the completion of the project.

4
 TABLE OF CONTENTS
CHAPTER NO. TITLE PAGE NO.

LIST OF FIGURES vi

LIST OF TABULATION vi

ABSTRACT vii

1 INTRODUCTION 7

2 LITERATURE SURVEY 8

3 PROBLEM STATEMENT AND OBJECTIVE 20

4 SCOPE OF THE PRESENT

INVESTIGATION
3.1 EXISTING SYSTEM 21

3.2 PROPOSED SYSTEM 21

5 HARDWARE COMPONENTS 26
5.1 AVR Atmega 328P 8-bit Controller 26
5.2 DC buck booster 14.1 volt 28
5.3 Voltage Sensor 29
5.4 Types of Voltage Sensors 30
5.5 L298N Motor Driver Module 37

6 CONCERNS 39
7 CONCLUSION AND SUMMARY
7.1 CONCLUSION 43
7.2 FUTURE SCOPE 43

8 REFERENCES 45

5
 ABSTRACT

E-bike is designed with Turbo boost BLDC Motors that can be operated at
higher speeds. Suspension systems must support both road holding/handling
and ride quality which are at odds with each other. The tuning of suspensions
involves finding the right compromise. It is important to control speed based
on the bumpy roads for the suspension to keep the road wheel in contact with
the road surface as much as possible because all the road or ground forces
acting on the vehicle do so through the contact patches of the tires. Our
Project, The Springless Suspension System Combines a differential
mechanism and an oscillating system that will support the E-bikes with relative
speed control in bumpy and terrain areas. A differential is a gear train with
three drive shafts that have a property such that the rotational speed of one
shaft is the average speed of the others or a fixed multiple of that average, and
Oscillation is the repetitive or periodic variation of an object. The Frame of the
system is mostly made of Mild steel. The frame that holds the tires oscillates
freely. The same frame is connected to the differential. A motor is connected
to a gear in the differential mechanism, thereby making it the driving gear. The
driving gear is connected by a bevel gear on both sides. Each gear is
connected to two wheels on each side. Each wheel is driven by an individual
motor. The motor makes sure that the gears and the wheels are in motion and
the oscillating property makes sure that the vehicle is moving forward even in
rough terrain. Through this Differential auto bevel mechanism speed control
with load balance on E-bikes can be maintained.

6
 CHAPTER 1
INTRODUCTION

We all are living in the future. With time, technology has improved our lives.
Nowadays, most countries are using electric bicycles. The use of cycles over vehicles
is always green for the environment, but an E-bike is the biggest adoption of green
transportation of the decade. Just think of E-bikes in place petrol operated scooters
rather than normal bikes. An E-bike uses rechargeable batteries that can travel up to
25 to 45 Kmph. As a result, it is faster than the normal cycle to reach your destination
quicker and in better shape. In this blog, we are going to learn about E-bike
technology, its types, and its advantages There are two main types of E-bikes- throttle
assist and pedal assist.

Fig 1.1 – Brief Model of E-Bike

An E-bike motor works by automatically switching on the (quiet) motor when you pedal
or throttle. A pedal-operated E-bike is the most popular option. As you pedal the bike,
the motor gets powered, and it works. In comparison, a throttle-assist E-bike is similar
to a normal motorbike. It operates as you accelerate the throttle. Electric bicycles are
available in many styles, from commuter bikes to full-suspension mountain bikes. The
power output of these pedal-operated motors is typically governed by regulations.

7
Mostly, they are available with an output power of around 250 watts. Bikes fitted with
a throttle-based motor system have slightly different output regulations. It can be
available with a maximum power of around 200 watts, while speed remains limited to
25 kmph.

CHAPTER 2

LITERATURE SURVEY

The following shows a survey done for an E-Bike speed control system based
on three modes of operation with an algorithm for real-time environmental
automation through Self-recording of data while operating in real-time Through
this brief survey, we have gained a huge layout of designing our system
architecture. Most of the content we gathered here is based on artificial
intelligence for Real-time traffic control system

Kousuke Mouri, Huimin Lu, Joo Kooi Tan, and Hyoungscop Kim, IEEE
Transactions on Machine Learning, 2021

Currently, in Japan, since the declining birth-rate and the aging of society
progresses, there is concern about the labor shortage in the productive age
population. The production age population has continued to increase after the
war, reaching 87.26 million in 1995. However, it decreased to 75.28 million in
2015, it is estimated that the population will be decreased to 45.29 million in
2065. There is a trend to compensate for labor shortage due to the decrease
in the productive age population by automation of factories using robots. One
of the industries in which the labor shortage is getting worse is the logistics
industry. The cause of the labor shortage in the logistics industry is that the
market size of ecommerce has expanded, in addition to the decrease in the
productive age population due to the declining birth-rate and aging population.
Furthermore, due to the expansion of the market size of ecommerce, there is
an increasing need for diversification of items ordered by consumers and
shortening of product delivery time, so it is necessary to improve work
efficiency. This paper aims to develop an automatic ordering product for
various products, managing inventory, stocking inventory, picking, packing,
delivering sorting, and delivering by robots. Picking is the operation of picking

8
up shelved items and storing them exactly in the box on the line. The cause of
difficulty in automating picking in the logistics industry is that it is necessary to
work on a large variety of products under different conditions, rather than often
working on the same type of objects repeatedly under the same conditions.
Therefore, by developing a picking method for various kinds of objects by the
robot, an automatic picking system expect to supplement the labor shortage in
the logistics industry and efficiently automate the flow from product arrival to
shipment. In this paper, in order to automate the picking operation using the
robot, we developed detection and recognition of the target object and
extraction method of the object region on a video image. The organization of
this paper is organized as the following. Section 2 describes the proposed
method of image processing for object detection and object extraction.

Pros:

Grow Cut is an interactive segmentation algorithm implemented in this project

that uses cellular automation as an image model. The automata evolution
model segmentation process was achieved with a 90% fast execution duty
cycle.

Cons:

Due to automata evolution, it vaguely draws some strokes inside the object
which reduces the system execution

Jiazhen Guo, Pen Chen, Yinlai Jiang, Hiroshi Yokoi and Shunta Togo,
IEEE Transactions on Global Conference on Life Sciences and
Technologies, 2021

With the development of augmented reality (AR) technology, human-robot

interaction (HRI) is becoming easier and easier. The wide application of AR in
HRI enhances the experience of humans interacting with robot systems or
wearable devices. The most representative product in the AR field today is
undoubtedly mixed reality device (Microsoft HoloLens). The use of HoloLens
in the robot is becoming more and more widely studied. However, using
HoloLens as robot vision for real-time object detection has been an unsolved
problem. Bahri et al. used TCP/IP internet connection and the system can only

9
reach 5 fps to detect all objects through HoloLens. In this work, we propose a
method to transmit real-time video on the client-side via HoloLens and apply
the YOLO model as a deep learning algorithm on the server-side to process
the client data.

Pros:

Implementation of Yolo process frame at the rate of 45 fps to 150 fps which is
better than real-time.

Cons:

Comparatively low recall and more localization error compared to faster

R_CNN which drastically pulls the system efficiency.

Toshihiro Yukawa, and Hideharu Okano, IEEE Transactions on Robotics

and Automation, 2014

In this study, a robot which detects dangerous articles made of plastic is

proposed. The principle of detection is to acquire information regarding the
difference of the electrification characteristics of plastic and its surroundings.
In an environment in which dangerous articles made of plastic have been
scattered, the robot searches for the dangerous articles using the sensor.
Previous methods for the detection of dangerous objects have included the
use of radar, magnetism, and magnetic flux, etc. The radar involved obtaining
information regarding a buried object, based on the relationship between an
incidence wave and the reflection wave of the electric wave. The principle of
this method is explained as follows. First, the electric wave is projected from
an antenna in all directions underground. The speed of the return of the
reflected wave will increase as the antenna approaches the buried object, with
the reflected wave becoming the fastest when the antenna is directly above
the object. The attainment time of this reflected wave can be measured. If the
density of the soil is known, then the speed of the electric wave through the
soil can be determined. The depth of where the object has been buried can be
obtained based on these factors. However, it is actually difficult to detect a
small object because soil density is not uniform. The magnetic method seeks

10
to locate into an underground object by using the phenomenon in which
terrestrial magnetism falls into disorder due to the magnetism of an
underground object. However, because the size of the magnetic field made by
the object is far smaller than terrestrial magnetism, it is considerably difficult
for this method to recognize an object of several centimetres. The metallic
detector is a device which uses the phenomenon of a whirlpool current.
Another magnetic pulse generated on the surface of a metallic object is
detected by the magnetic flux of the search coil. The magnetic flux response
alters the electromotive power of the search coil. In other words, a metallic
detector recognizes metal by detecting a change in electromotive power. The
above techniques have been used by many researchers in the development of
robots intended to detect the location of land mines etc. In contrast to these
methods, the method described in this paper, in which electrification is
employed, can be used to detect dangerous plastic objects concealed
underground. (i.e., under sand or soil). As previously described, the principle
in which dangerous articles made of plastic can be detected involves
distinguishing plastic and other materials by measuring the surface potential of
the object after the object to be detected is electrified. Robots employing this
technique can search for the plastic articles by using a new sensor, and
achieving the movement using two wheels and the movements of an arm. This
new detection sensor combines a corona charger, a surface potential sensor,
and a distance sensor. Next, we explain kinds of dangerous articles made of
plastic. For instance, bombs, land mines, and plastic containers (UN standard
acceptable goods) in which dangerous articles can be hidden. Plastic
containers also include a medium container made of hard plastic, a small
plastic drum, a box made of plastic, and a plastic bag, etc. Another method for
the robotic detection of plastic objects involves the use of visual information
acquired by a camera. However, this method encounters some difficulty in
accurately identifying the images of the objects acquired. Other electronic
equipment which has employed the electrification phenomenon have included
electrostatic micromachines, a hair transplant device, and the electronic
photograph.

Pros:

11
Picturization and false gain rate is reduced to 20% when compared to
rainforest algorithm in the paper as discussed.

Cons:

The rate of algorithm training is highly expensive since and execution speed is
very low due to higher bandwidth

Yeong-Hwa Chang, Ping-Lun Chung, and Hung-Wei Lin ,IEEE

Transactions on Applied System Innovation, 2018

Traditional service robots are usually lack of cognitive services such as image
and voice identification. There still exists a clear gap how to integrate service
robots with human intelligence. Recently, machine learning techniques have
been widely discussed. There are many remarkable progresses that can be
used to fulfil cognitive services. Deep learning algorithms have been popularly
addressed that may lead to a new generation in machine learning discipline.
For example, a deep learning neural network was applied for hand written
recognition. In recently years, deep learning methods have emerged as
powerful machine learning methods for object recognition and detection.
Compared with traditional approached, deep learning methods can quickly
learn the features directly from the raw images. In deep learning algorithms,
Faster R-CNN (Region Convolution Neural Network) is one newly developed
algorithm. According to different datasets, Faster R-CNN was implemented for
object detection. Object recognition and detection is a key problem in the area
of cognitive service, especially in moving objects. Traditional object recognition
methods basically focused on improving recognition performance on specific
datasets. In general, PASCAL VOC, Microsoft COCO, and ImageNet are three
popularly mentioned datasets. These datasets provide sufficient variability in
object categories and instances, the training data mostly consists of images of
arbitrarily picked scenes and objects. If mobile robots can integrate with deep
learning-based object detection, more interesting applications could be
generated. Robot Operating System (ROS) is a robotic middleware with many
open resource packages. It will be more efficient to develop applications using
a ROS framework.

12
 In the control of ROS-based mobile robot was addressed using Arduino
platform. In past few years, mobile robot integrated with deep learning method
based on ROS middleware has attracted a lot of attention, such as the
applications on shopping, navigation, and environment recognition. In this
paper, object detection and recognition can also be combined with mobile
robot which is setting on ROS middleware. Some ROS packages are designed
for the mobile robot to attain the desired tasks, such as robot control and
simultaneously localizing and mapping. Based on Raspberry Pi, an easy-to-
use control kernel is developed to integrate frontend and back-end
applications. In front-end part, the raw images captured by the Kinect sensor
are provided for object detection by using Faster R-CNN algorithm and GPU
accelerated computing.

Pros:

This project proposal features the whole system to work independently on

automatic deduced and optically tuned outcome. Neural network-based
approach can also be reliable in many different stages of data processing.

Cons:

It requires very large amount of data in order to perform better than other
techniques. The training algorithm is highly expensive and complex data
models.

Donghoon Kim, Donghwa Lee, and Hyun Myung, IEEE Transactions on

Applied System Innovation, 2018

Navigation techniques in underwater environment are important research

themes in developing autonomous underwater robots. However, compared to
the ground, the underwater environment is highly dynamic and has limited
available communication and sensors, as well as the sensors have different
characteristics due to the medium. Thus, underwater navigation techniques
such as localization, path planning and following are challenging problems but
essential to successful performance. The techniques have been studied to
complete tasks including docking station cable tracking for inspection purpose,
and underwater area coverage for monitoring purpose. Vision data through a
camera is one of useful information for autonomous navigation. Since the

13
visual data provide detailed information about environment at low cost, the
data are used for the human as well as autonomous robot’s navigation. Thus,
although there are disadvantages of limited detection range and difficulty of
feature extraction for cameras, the visual data still very useful for close range
detection. Many vision-based researches for the underwater robots. Artificial
objects used the feature points and the color information of the environment,
the color distribution and shapes of artificial landmarks for. And the usefulness
for AUV’s vision-based docking was demonstrated by detecting or tracking of
artificial landmarks using LEDs, Self-Similar pattern boards and Haar-like
rectangular features. In other studies, computer vision techniques have been
used for underwater cable tracking and inspection.

In this paper, some vision-based techniques for autonomous navigation

of the underwater robots were proposed. First, in order to implement computer
vision techniques in noisy underwater environment, the pre-processing
methods were introduced. Next, template matching-based object detection and
mean shift tracking algorithm were introduced. But, since the detection method
was sensitive to the illumination changes or noise, we proposed a novel
template matching algorithm based on weighted correlation coefficient and
multiple template-based selection. Finally, the experimental results show that
the proposed algorithm demonstrates satisfactory detection performance. The
paper is organized as follows. Section II describes object detection and
tracking algorithms including the artificial underwater objects, methods and
effects of the pre-processing, conventional and proposed template matching,
and mean-shift tracking. And the environments, conditions, and results of
experiments are represented

Pros:

Aquatic vision-based detection is a novel thought process that support the

entire project flow. The database maintained here is based on open cloud
access which can be retrieved on any situation

Cons:

Due to open source this system can be claimed for patent ship and can be
modified easily

14
Wang Canyong, IEEE Transactions on robotics and Intelligent Systems,
2018

Traffic signs occupy an important position in the road traffic system. The main
function of the traffic signs is to display the contents that need to be noticed in the
current road section, to prompt the drivers in front of the road the danger and
difficulty in the environment, to warn the driver to drive at the prescribed speed, to
provide a favorable guarantee for safe driving. Therefore, the detection and
identification of traffic signs is very important research direction, which is of great
significance to prevent road traffic accidents and protect the personal safety of
drivers. Road traffic signs are divided into two major categories of main signs and
auxiliary signs, mandatory signs, guide signs, tourist signs and road construction
safety signs. among them, prohibition signs mainly played a role in banning
certain kinds of behavior, a total of 43 categories. mandatory signs indicate the
role of vehicles, pedestrians, a total of 29 categories, mandatory signs mainly
placed in the need to indicate vehicles, pedestrian’s road sections or near the
intersection. Warning signs are mainly to alert drivers, vehicles and pedestrians
to beware of dangerous targets, a total of 45 categoriesThe realistic application
scenarios bring high reliability requirements for traffic sign detection and
identification. If there is a misdetection or a misdetection, or a misidentification
detected afterwards, it will result in erroneous information being input to the
vehicle, causing erroneous operation and unpredictable consequences.
Therefore, it is unavoidable to develop a traffic sign detection and recognition
algorithm with high accuracy and robustness under various complicated
situations and disturbances.

Pros:

Features are automatically deduced and optimally tuned for desired

outcome. This avoids time consuming machine learning techniques

Cons:

It requires very large amount of data in order to perform better than other
techniques. Which is extremely expensive to train due to complex data models.

15
Zhao Boxin, Jiang Jun, Niu Yifeng, and Sheng Lincheng, IEEE
Transactions on Intelligent System Design and Engineering
Applications, 2012

Recent years have seen numerous research efforts studying the target
tracking. From a robotic vehicle perspective, this research is widely used in
various domains, including robot vision navigation, intelligent monitoring, and
motion planning. In an urban environment, traffic sign tracking which continues
to pose significant research challenges, is not only the basic step of
understanding environment, but an important research content of Vision
Navigation, which requires real-time identification and decision-making, as well
as optimization of many parameters (including taking into account the physical
constraints of the vehicle). With high speed and robustness to rotation and
deformation of the object, Mean-Shift plays an important role in target tracking.
The Mean shift algorithm has the characteristics of nonparametric and fast
pattern matching, but because of lacking the necessary updates, the algorithm
easily causes loss of target when the target size changes significantly. To
improve these, Cam-Shift stands out as the fastest and simplest. But Cam-
Shift often fails tracking when the color information of the target region is similar
with the background. Recently, many improved Cam-Shift Algorithms have
been presented. New features including improved LBP, color information is
introduced. In Cam-shift and Kalman algorithms are combined to improve the
matching accuracy of the target. This paper designs a traffic sign probability
model, and presents a depth-first region search algorithm based on the
probability. This method concentrates to remove the redundant background
information, and detect/tracking the traffic sign by using Cam-shift algorithm.

Pros:

This system developed a universal image processing model which can be

applied to many different applications and data types.

Cons:

There is not standard theory to guide in selecting right learning process tools
as it requires knowledge of topology, training method and other parameters.

16
Tiago Moura, Antonio Valente, Antonio Sousa and Vitor Filipe , ICARSC
Transactions on Autonomous Robot Systems and Competitions, 2021

Robotic contests are one of the most promising ways to attract students to the
field of robotics, since winning an award at a competition not only gives
students a sense of accomplishment but also gives pride and visibility to
schools. Autonomous Driving is one of the seniors (manly for university
students) contests in the National Robotic Festival. The Autonomous Driving
Competition is a contest for fully autonomous robots that takes place in a track
with the shape of a traffic road, surrounded by two parallel side lines and
including two lanes separated by a dashed mid line, depicted on Fig. 1a. In the
case of advanced robots (Challenge Class) there are different traffic signs
(warning signs, mandatory signs, and information or services signs) that must
be detected and recognized, there are also traffic lights mounted right above
the zebra crossing (inverted 17’ TFT monitors) in order to conduct the robot
trials and giving them orders to STOP, follow straight ahead, turn left, parking,
and end of trial. Since 2011 University of Trans-Montes and Alto Douro
participates on this competition with a four-wheel Ackermann steering
geometry robot (UTAD McQueen). The robot chassis is from a 1/5 scale RC
car racing adapted with a Maxon EC brushless DC motor droved by a Maxon
EPOS 70/10 4-Q servo controller and the steering is controlled by a servo
motor. All the low-level control (steering, Maxon EPOS, light, and RGB LED to
tell the jury the traffic sign detected) is performed by an Arduino Mega 2560
board. The main controller is a laptop (Samsung RC530, Core i7, with 6 GB of
RAM), which communicates with the Arduino board through a USB serial port.

Pros:

This project proposal features the whole system to work independently on

automatic deduced and optically tuned outcome. Neural network-based
approach can also be reliable in many different stages of data processing.

Cons:

17
It requires very large amount of data in order to perform better than other
techniques. The training algorithm is highly expensive and complex data
models.

Mohammed Ikhlayel, Adre Johan Isware, Arief Kurniawan, Ahmad Zaini

and Eko Mulyanto Yuniarno, CENIM Transactions on Network and
Intelligence Media, 2020

Driving is an activity with many variations in decision making and action. Based
on the driver’s effort, driving activities are divided into three tasks: strategic
effort, tactical effort and operational effort. Strategic effort involving trip
planning, such as determining when to travel, how to travel, and determining
the best route. The tactical effort involves manoeuvring vehicles in traffic such
as changing vehicle lanes, choosing the appropriate vehicle speed, and
overtaking other vehicles. On the tactical effort, the driver must be monitoring
driving environment by detecting, recognizing, classifying objects and events
on the road. An appropriate response, such as overtaking and detecting other
vehicles, is provided based on this information. Operational effort that involves
a quick reaction to the vehicle’s control, such as braking and accelerating the
vehicle’s speed, driving corrections to take the vehicle stays on track. An
autonomous car must be able to provide these three tasks automatically
without human assistance. One of the essential features in an autonomous
vehicle is the ability to recognize traffic signs. Autonomous cars detect and
classify traffic signs found on the road and help the drivers to navigate follow
the directions given by these traffic signs. In this research, we developed a
prototype of self-driving Autonomous Car based on traffic sign information. The
Autonomous Car prototype uses cameras to detect traffic signs on the road in
real-time. YOLO’s Deep Learning Algorithm is employed to classify detected
traffic signs. The result is fed to the autonomous vehicle to provide autonomous
response according to detected traffic signal commands.

Pros:

18
Picturization and false gain rate is reduced to 20% when compared to
rainforest algorithm in the paper as discussed.

Cons:

Rate of algorithm training is highly expensive since and execution speed is

very low due to higher bandwidth

Wahyono Laksono Kurnianggoro and Kang-Hyun Jo,SICE Transactions

on Artificial Neural Networks, 2021

Traffic signs, as one of the important information for autonomous vehicle, are
used for traffic warning, regulation, routing and management. These signs are
intended to affect the behavior of drivers. Hence, traffic sign recognition (TSR)
system plays an important role for autonomous navigation systems. In the past
two decades, many approaches of traffic sign recognition have been proposed.
However, based on our survey, mainly, most of them consist of three main
stages: Region segmentation for extracting the candidate regions. In this
stage, color feature is usually exploited due to specific color of the sign, Shape
analysis for classifying signs into circle, rectangle and triangle shape, and the
last Recognition for identifying the sign as certain class and its meaning where
previously feature extraction is performed. Specifically, the recognizing step is
done by several classification techniques. Artificial neural network (ANN), k-
nearest neighbour (KNN), support vector machine (SVM), and random forest
(RF) are among the most commonly used. Reference reported that, KNN is
the best classifier among these classification techniques. However, it takes
longest time for classification and recognition process. Tracking by detection
is one of the most common methods for handling this kind of problem. Mostly,
the tracking method can reduce the false negative and positive as the true
traffic sign should appear in constantly in several frames.

Pros:

Features are automatically deduced and optimally tuned for desired

outcome. This avoids time consuming machine learning techniques

Cons:

It requires very large amount of data in order to perform better than other

19
techniques. Which is extremely expensive to train due to complex data models.

CHAPTER 3

PROBLEM STATEMENT AND OBJECTIVES

E-bike requires E stations that must be maintained with a proper power source.
GPS tracking system must be installed in all cases of the units to solve the
connectivity issues in case of battery failure. This would also increase the
productivity of the battery cells which exhaust the natural resource of
renewable energy. In the case of energy service these bikes must be taken
under proper tech persons to sort the problems. These kinds of non-gear bikes
can only be designed with BLDC motor which is expensive in the market.

To auto-control the speed of E-bikes through differential bevel gears. To

increase the system suspension differential balance between the cob joint
during speed control. To increase friction activity during sudden speed control.
To main the data log of the road activity during speed control on the regular
road study. To analyze the behavior and physical parameter of the road to
initiate turbo start.

20
 CHAPTER 4

EXISTING SYSTEM AND PROPOSED METHODOLOGY

EXISTING SYSTEM
No Proper Battery Suge current calculations are made during Mode switching.
Speed control is manual not on sensor data. Data log is not maintained for
battery service and fuel ability calculation. CAN-based centralized controllers
are not specially designed which is even our projects' future scope to handle
day-to-day data. GPS tracking is not powered by free energy we can modulate
the mode speed to recharge the system based on the reverse suspension
effect.

PROPOSED SYSTEM
Our Proposed model is based on the active and effective use of an Electrical
power source over the animals through bob wires. Indications are set to
eradicate the intrusions. In the initial stage, PIR motion sensor activity control
is set. Followed by a high-decibel siren to express the wild intrusions and
thieves’ activity. A proper IoT-connected network will indicate the notification
to the farmers regarding the activity. Finally, the electric bob wire will be
activated. Here we have reduced 96% of data loss when compared to all
traditional models and systems. We propose a system that not only responds
to the activity further it can print out the statement too. The logic factor is stable
and precise when compared to academics. DF mini can handle up to 32 GB of
memory decoding unit, here we can set our command lines to accomplish the
project.

21
BLOCK DIAGRAM:

Fig: BLOCK DIAGRAM

Algorithm of Our Working Model:

PHASE 1 – E-Bike BLDC PWM Tooth Generator

• This Phase deals with the BLDC Rotation control from the bevel connectors
• Based on the Duty cycles released from the throttle the modes can be
changed into three operations such Normal mode, Economy, and Sports
Mode.
• Based on this mode selection BLDC will be coupled with the internal
Centrifugal load transmitters
• Without proper PWM signals mode selection is not possible to rotate the
gears
• As per our approximation system speed varies from 30kmp – 120kmp

22
PHASE 2 – Vibration Sensor Activity Log Calculator

• Based on the Road Condition data log will be generated for a safe ride.
• Sport mode will not be activated in the vibration data from the CAN
bus is high.
• This prominently helps us to increase the stability of the system.

PHASE 3 – Battery Charge Cut-off

• On Manual tripping this engine system can attain full load and full speed
• Bevel transmissions will not take into data activity
• In such cases high usage of the battery will reduce the battery life and create
a lot of system failures
• So, it is mandated to check the battery drain rate based on the consumption
ratio

Advantages of the Proposed System

Upcoming key points are the current drawbacks that still exist in the following
methods. In our proposal, we came across sorting out the listed errors
1. Segregation process is only implemented in our system
2. Bundling and package part will be developed in our future plans
3. Monitoring of data on the cloud required the power of the internet and
separated server space
4. But when compared to the entire profit of the project this system loss is
negligible
5. 24x7 power backup must be maintained for proper updates

23
In gear transmissions, vibration causes noise and malfunction. In actual
applications, misalignments contribute to intensifying the destructive effect of
vibrations. In this project, the nonlinear dynamics of a spiral bevel gear pair,
with a small helix angle, considering different misalignments, are deeply
investigated. Axial misalignment, radial misalignment, and the combination of
these two types are considered in this study. The governing equation is
numerically solved through an implicit Runge–Kutta scheme. Since the main
goal of this study is the analysis of the dynamic scenario, the mesh stiffness of
the gear pair is obtained from the literature. The dynamical system is nonlinear
and time-varying; it is analyzed through time responses, phase portraits,
Poincare maps, and bifurcation diagrams. Results show that among the
considered three cases with different types of misalignments, the spiral bevel
gear with axial misalignment is the worst destructive case; aperiodic,
subharmonic, and multiperiod responses are observable for this case. It is
interesting that the chaotic responses for the case, having both types of
misalignments, are less likely for the case with axial misalignment. These are
the list of alternatives that must be changed to control the speed of BLDC
motors based on the Physical Road conditions.

Fig 5.1 – Architecture of Mode Switching on E-Bike

24
Hardware Required

• AVR Atmega 328P 8-bit Controller

• DC Buck booster 14.1 volt
• Battery Swap 1,2
• Voltage Sensor 1,2
• 16*2 LCD Display
• PCF8574A 12C Module
• Wireless Bluetooth Device
• L298N Motor Driver Module
• Solar Panel 25v
• 12v Rechargeable Batteries

25
 CHAPTER 5

HARDWARE COMPONENTS

AVR Atmega 328P 8-bit Controller:

ATmega328 8-Bit AVR MCUs are low-power, CMOS 8-bit microcontrollers based on
the AVR® enhanced RISC architecture. The devices execute instructions in a single
clock cycle to achieve CPU throughput approaching one million instructions per
second (MIPS) per megahertz. This allows for optimizing power consumption versus
processing speed.

Features:

• High performance, low power AVR® 8-Bit Microcontroller family

• Advanced RISC architecture
o 131 powerful instructions – most single clock cycle execution
o 32 x 8 general-purpose working registers
o Fully static operation
o Up to 20 MIPS throughput at 20MHz
o On-chip 2-cycle multiplier
• High endurance non-volatile memory segments
o 4/8/16/32KBytes of in-system self-programmable Flash program memory
o 256/512/512/1KBytes EEPROM
o 512/1K/1K/2KBytes internal SRAM
o Write/erase cycles 10,000 Flash/100,000 EEPROM
o Data retention 20 years at 85°C/100 years at 25°C(1)
o Optional boot code section with independent lock bits
• In-system programming by an on-chip boot program
• True read-while-write operation
o Programming lock for software security
• QTouch® library support
o Capacitive touch buttons, sliders, and wheels
o QTouch and QMatrix™ acquisition
o Up to 64 sense channels

26
• Peripheral features
o Two 8-bit timer/counters with separate prescaler and compare mode
o One 16-bit timer/counter with separate prescaler, compare mode, and
capture mode
• Real-time counter with separate oscillator
o Six PWM channels
o 8-channel 10-bit ADC in TQFP and VQFN package
• Temperature measurement
o 6-channel 10-bit ADC in SPDIP package

• Temperature measurement
o Programmable serial USART
o Master/slave SPI serial interface
o Byte-oriented 2-wire serial interface (Philips I2C compatible)
o Programmable watchdog timer with separate on-chip oscillator
o On-chip analog comparator
o Interrupt and wake up on Pin change
• Special microcontroller features
o Power-on reset and programmable brown-out detection
o Internal calibrated oscillator
o External and internal interrupt sources
o Six sleep modes: Idle, ADC Noise Reduction, Power-save, Power-down,
Standby, and Extended Standby
• I/O and packages
o 23 programmable I/O lines
o 28-pin SPDIP, 32-lead TQFP, 28-pad VQFN, and 32-pad VQFN
• Operating voltage
o 1.8 - 5.5V
• Temperature range
o -40°C to 85°C
• Speed grade
o 0-4MHz at 1.8-5.5V, 0-10MHz at 2.7-5.5.V, 0-20MHz at 4.5-5.5V

27
 • Power consumption at 1MHz, 1.8V, 25°C
o Active mode 0.2mA
o Power-down mode 0.1µA
• Power-save mode 0.75µA (including 32kHz RTC)

Fig 5.1 ATmega328p

DC buck booster 14.1 volt

The buck–boost converter is a type of DC to DC Converter that has an output

voltage magnitude that is either greater than or less than the input voltage
magnitude. It is equivalent to a flyback converter using a single inductor instead of
a transformer. Two different topologies are called buck–boost converter. Both of
them can produce a range of output voltages, ranging from much larger (in absolute
magnitude) than the input voltage, down to almost zero.

In the inverting topology, the output voltage is of the opposite polarity than the input.
This is a switch mode power supply with a similar circuit topology to the boost
converter and the buck converter. The output voltage is adjustable based on the of
the switching duty cycle transistor. One possible drawback of this converter is that
the switch does not have a terminal at ground; this complicates the driving circuitry.
However, this drawback is of no consequence if the power supply is isolated from
the load circuit (if, for example, the supply is a battery) because the supply and diode
polarity can simply be reversed. When they can be reversed, the switch can be on
either the ground side or the supply side.

28
When a buck (stepdown) converter is combined with a boost (stepup) converter the
output voltage is typically of the same polarity of the input, and can be lower or
higher than the input. Such a non-inverting buck-boost converter may use a single
inductor which is used for both the buck inductor mode and the boost inductor mode,
using switches instead of diodes, sometimes called a "four-switch buck-boost
converter, it may use multiple inductors but only a single switch as in the SEPIC
and Cuk topologies.

Fig 5.2 Buck boost Converter

Voltage Sensor

Generally, a sensor is an electrical device used to detect as well as respond to a

particular type of signal like optical or electrical. Implementation of sensor
techniques in voltage or current has become an outstanding option toward the
measurement of voltage & current methods. The advantages of sensors over
conventional methods for measuring mainly include less size and weight, high
safety, high accuracy, non-saturable, eco-friendly, etc. It is feasible to merge both
the current and voltage measurement into a physical device with tiny and solid
dimensions. This article discusses an overview of the voltage sensor and its
working.

• This sensor is used to monitor, calculate and determine the voltage supply.
This sensor can determine the AC or DC voltage level. The input of this
sensor can be the voltage whereas the output is the switches, analog voltage
29
 signal, a current signal, an audible signal, etc. Some sensors provide sine
waveforms or pulse waveforms like output & others can generate outputs
like AM (Amplitude Modulation), PWM (Pulse Width Modulation) or FM
(Frequency Modulation). The measurement of these sensors can depend on
the voltage divider.
• This sensor includes input and output. The input side mainly includes two
pins namely positive and negative pins. The two pins of the device can be
connected to the positive & negative pins of the sensor. The device positive
& negative pins can be connected to the positive & negative pins of the
sensor. The output of this sensor mainly includes supply voltage (Vcc),
ground (GND), analog o/p data

Types of Voltage Sensors

These sensors are classified into two types like a resistive type sensor and
capacitive type sensor.

1) Resistive Type Sensor

This sensor mainly includes two circuits like a voltage divider & bridge circuit. The
resistor in the circuit works as a sensing element. The voltage can be separated into
two resistors like a reference voltage & variable resistor to make a circuit of the
voltage divider. A voltage supply is applied to this circuit. The output voltage can be
decided by the resistance used in the circuit. So the voltage change can be
amplified.

30
2) Capacitor Type Sensor

This type of sensor consists of an insulator and two conductors within the center.
As the capacitor is power-driven with 5 Volt, then the flow of current will be there in
the capacitor. This can create revulsion of electrons within the capacitor. The
difference in capacitance indicates the voltage and the capacitor can be connected
within the series.

Fig 5.3 Voltage Detection Sensor

16x2 LCD Display

16×2 LCD is named so because; it has 16 Columns and 2 Rows. There are a lot of
combinations available like, 8×1, 8×2, 10×2, 16×1, etc. But the most used one is the
16*2 LCD, hence we are using it here.

31
All the above mentioned LCD display will have 16 Pins and the programming
approach is also the same and hence the choice is left to you. Below is the Pinout
and Pin Description of 16x2 LCD Module:

The LCD can work in two different modes, namely the 4-bit mode and the 8-bit
mode. In 4 bit mode we send the data nibble by nibble, first upper nibble and then
lower nibble. For those of you who don’t know what a nibble is: a nibble is a group
of four bits, so the lower four bits (D0-D3) of a byte form the lower nibble while the
upper four bits (D4-D7) of a byte form the higher nibble. This enables us to send 8
bit data.

Whereas in 8 bit mode we can send the 8-bit data directly in one stroke since we
use all the 8 data lines.

Now you must have guessed it, Yes 8-bit mode is faster and flawless than 4-bit
mode. But the major drawback is that it needs 8 data lines connected to the
microcontroller. This will make us run out of I/O pins on our MCU, so 4-bit mode is
widely used. No control pins are used to set these modes. It's just the way of
programming that change.

Read and Write Mode of LCD:

As said, the LCD itself consists of an Interface IC. The MCU can either read or write
to this interface IC. Most of the times we will be just writing to the IC, since reading
will make it more complex and such scenarios are very rare. Information like position
of cursor, status completion interrupts etc. can be read if required, but it is out of the
scope of this tutorial.

32
 Fig 5.4 16x2LCD Display

PCF8574A 12C Module

KEY FEATURES OF PCF8574 I2C I/O EXPANSION MODULE:

• 8 bi-directional data lines

• Loop-thru feature allows expansion of up to 8 modules / 64 data lines
• I2C interface with jumper adjustable addresses
• Interrupt output capability
• 3.3V and 5V compatible.

A common requirement when working with MCUs is the need to add more digital
I/O than the device supports natively. The PCF8574 is one of the more popular
methods of adding lines as it uses the I2C bus that requires only 2 lines on the
MCU. It provides 8 additional digital I/O lines which are easily expandable up to 64.

I2C Interface

The base I2C address of the modules is 0x20.

There are three address jumps (A0-A2) the determines which I2C address to
use. As shipped, these jumpers are all set to the ‘-‘ side which is ground or LOW as
shown in the picture. The ‘+’ side is Vcc or HIGH.

33
This puts the module at the base address of 0x20. The jumpers can be moved in a
binary fashion to increase the address, so the address can range from 0x20 to 0x27
as shown in the table below.

If you daisy-chain the modules, you will need to set a different address for each of
the modules.

Address A2 A1 A0
(Hex)

0x20 LOW LOW LOW

0x21 LOW LOW HIGH

0x22 LOW HIGH LOW

0x23 LOW HIGH HIGH

0x24 HIGH LOW LOW

0x25 HIGH LOW HIGH

0x26 HIGH HIGH LOW

0x27 HIGH HIGH HIGH

You may run into devices with the PCF8574A part installed. If that happens, don’t
panic. These just use a different starting I2C address of 0x38. This part is offered
by the mfr so that if both A and non-A parts are used together in a system, the

34
number of modules can be increased up to a maximum of 16 providing at total of
128 digital lines. The ‘T‘ marking on the devices just denote that it is a surface
mount device.

If there is ever a doubt about the I2C address of this or any device, just hook it up
to the I2C bus and apply power and ground and then run the I2C scanner software.

1.1.1 I/O Functionality

The I/O is defined as quasi-bidirectional. A quasi-bidirectional I/O is either an input

or output port without using a direction control register. When set as inputs, the pins
act as normal inputs do. When set as outputs, the PCF8574 device drives the
outputs LOW with up to 25mA sink capability but when driving the outputs HIGH,
they are just pulled up high with a weak internal pull-up. That enables an external
device to overpower the pin and drive it LOW.

The device powers up with the 8 data lines all set as inputs.

When using the pins as inputs, the pins are set to HIGH by the MCU, which turns
on a weak 100 uA internal pull-up to Vcc. They will read as HIGH if there is no input
or if the pin is being driven HIGH by an external signal but can be driven LOW by
an external signal that can easily override the weak pull-up.

If used as outputs, they can be driven LOW by the MCU by writing a LOW to that
pin. A strong pull-down is turned on and stays on to keep the pin pulled LOW. If the
pin is driven HIGH by the MCU, a strong pull-up is turned on for a short time to
quickly pull the pin HIGH and then the weak 100uA pull-up is turned back on to keep
the pin HIGH.

If the pins are set to be outputs and are driven LOW, it is important that an external
signal does not also try to drive it HIGH or excessive current may flow and damage
the part.

Whenever the internal register is read, the value returned depends on the actual
voltage or status of the pin.

35
The I/O ports are entirely independent of each other, but they are controlled by the
same read or write data byte.

Interrupt Output

The interrupt open drain output pin is active LOW. It is normally pulled HIGH using
a pull-up resistor and is driven low by the PCF8574 when any of the inputs change
state. This signals the MCU to poll the part to see what is going on. If connecting
this pin, enable the internal pull-up resistor on the MCU or add an external pull-up
of 10K or so.

If using interrupts with multiple modules, since they are open drain they can be tied
together if a single interrupt back to the MCU is desired.

Module Connections

The connections to the module are straight forward.

1. Supply 3.3 or 5V power and ground.

2. Connect I2C SCL and SDA lines to same on the MCU.
3. If used, connect the INT line to an interrupt input on the MCU and use a pull-
up resistor.

1 x 4 Header (Male & Female)

• VCC = Vcc (2.5V – 6V) typically connects to uC 3.3V or 5V

• GND = Ground connects to uC ground
• SDA = I2C SDA connects to uC I2C SDA
• SCL = I2C SCL connects to uC I2C SCL

1 x 9 Header

• P0 = I/O Pin 0
• P1 = I/O Pin 1
• P2 = I/O Pin 2
• P3 = I/O Pin 3
• P4 = I/O Pin 4

36
 • P5 = I/O Pin 5
• P6 = I/O Pin 6
• P7 = I/O Pin 7
• INT = Interrupt Output – connects to MCU interrupt pin. Active LOW

Fig 5.5 PC8574A 12C Module

L298N Motor Driver Module

This L298N Motor Driver Module is a high power motor driver module for
driving DC and Stepper Motors. This module consists of an L298 motor driver
IC and a 78M05 5V regulator. L298N Module can control up to 4 DC motors,
or 2 DC motors with directional and speed control.

L298N Module Pinout Configuration

Pin Name Description

IN1 & IN2 Motor A input pins. Used to control the spinning direction of
Motor A

IN3 & IN4 Motor B input pins. Used to control the spinning direction of
Motor B

37
 ENA Enables PWM signal for Motor A

ENB Enables PWM signal for Motor B

OUT1 & Output pins of Motor A

OUT2

OUT3 & Output pins of Motor B

OUT4

12V 12V input from DC power Source

5V Supplies power for the switching logic circuitry inside L298N IC

GND Ground pin

Features & Specifications

• Driver Model: L298N 2A

• Driver Chip: Double H Bridge L298N
• Motor Supply Voltage (Maximum): 46V
• Motor Supply Current (Maximum): 2A
• Logic Voltage: 5V
• Driver Voltage: 5-35V
• Driver Current:2A
• Logical Current:0-36mA
• Maximum Power (W): 25W
• Current Sense for each motor
• Heatsink for better performance
• Power-On LED indicator

38
Alternate Driver Modules: TMC2209, DRV8825, A4988, L9110S, DRV8711

Related Components: LM298 Motor Driver IC, 78M05 Voltage

Regulator, Capacitors, Resistors, Heat Sink

Brief about L298N Module

The L298N Motor Driver module consists of an L298 Motor Driver IC, 78M05
Voltage Regulator, resistors, capacitor, Power LED, 5V jumper in an integrated
circuit.

Fig 5.5 L298N Motor Driver

78M05 Voltage regulator will be enabled only when the jumper is placed. When
the power supply is less than or equal to 12V, then the internal circuitry will be
powered by the voltage regulator and the 5V pin can be used as an output pin
to power the microcontroller. The jumper should not be placed when the power
supply is greater than 12V and separate 5V should be given through 5V
terminal to power the internal circuitry.

ENA & ENB pins are speed c ontrol pins for Motor A and Motor B while IN1&
IN2 and IN3 & IN4 are direction control pins for Motor A and Motor B.

39
40
41
 CHAPTER 6
CONCERNS

CHAPTER 7

ADVANTAGES

• Allows seniors or those physically unable to far travel longer distances.

• The downtube battery keeps weight low and Enhances balance.
• It is compact and requires very little maintenance.
• Allows you to travel faster than a regular bike(especially when going uphill).
• Easy to maintain, it's just an ordinary bike with an added battery and motor.
• Very useful for commuting quickly and avoiding traffic.
• If the battery is empty, you can just cycle normally, which may help you lose
weight.
• High mobility & versatility - overcomes off-road, dirt, snow.

DISADVANTAGES

• It's heavy, which makes it difficult to store and very difficult to pedal up a hill
if you run out of battery power
• Short battery life and long charging time, should you buy a 10 Ah 36 V battery,
you are likely limited to around 50 km (36 mi)
• Double the price of a standard bicycle
• Electric Bikes May Cost More Than Traditional Bicycles
• Electric Bikes Can Be Hefty
• Electric Bike Riding Rules Aren’t Always Clear
• Electric Bike Batteries Can Be a Hassle

42
 CHAPTER – 8
CONCLUSION
In conclusion, the E-bike speed control system is an essential component of
electric bicycles that helps riders maintain a safe and comfortable speed while
riding. The system employs various technologies, such as torque sensors, pedal-
assist sensors, and throttle controls, to regulate the motor's speed and power
output. The E-bike speed control system offers several benefits to riders,
including increased safety, improved battery life, and enhanced performance.
With the increasing popularity of electric bicycles, the development of more
advanced and efficient E-bike speed control systems is expected to continue,
improving the overall riding experience and making E-bikes an even more viable
alternative to traditional bicycles and motorized vehicles.

In addition to enhancing the overall riding experience, the E-bike speed control
system also plays a critical role in ensuring compliance with local regulations and
laws. The system enables manufacturers to set maximum speed limits for their
electric bicycles to meet the legal requirements of different regions. By adhering
to these speed limits, riders can avoid potential fines or legal consequences for
violating local laws. Moreover, the speed control system's ability to regulate the
motor's output makes E-bikes more accessible to a broader range of riders,
including those who may not have the strength or ability to ride a traditional
bicycle. As such, the E-bike speed control system represents a significant
advancement in personal transportation technology, providing a safer, more eco-
friendly, and efficient alternative to traditional modes of transportation.

43
 REFRENCEE
[1] T. Moura, J. Teixeira, F. Tuna, F. Moreira, A. Valente, V. Filipe, S. Soares.
“Reconhecimento de Sinais de Trânsito para-Prova de Robótica de Condução
Autónoma”. Proceedings of 19th Annual Seminar on Automation, Industrial
Electronics and Instrumentation (SAAEI’12), pp. 645-649, 2012.

[2] V. Prisacariu, R. Timofte, K. Zimmermann, I. Reid, and L. Van Gool,

“Integrating object detection with 3D tracking towards a better driver assistance
system”, in Proc. 20th ICPR, pp. 3344–3347, August 2010.

[3] V. Prisacariu, R. Timofte, K. Zimmermann, I. Reid, and L. Van Gool,

“Integrating object detection with 3D tracking towards a better driver assistance
system”, in Proc. 20th ICPR, pp. 3344–3347, August 2010.

44
[4] X. Qingsong, S. Juan, and L. Tiantian, “A detection and recognition method
for prohibition traffic signs”, in Proc. Int. Conf. IASP, pp. 583–586, April 2010.

[5] F. Ren, J. Huang, R. Jiang, and R. Klette, “General traffic sign recognition
by feature matching”, in Proc. 24th Int. Conf. IVCNZ, pp. 409– 414, November
2009.

[6] E. Krsak, S. Toth, “Traffic sign recognition and localization for databases of
traffic signs”, in Acta Electrotechnica et Informatica, vol. 11, no. 4, pp. 31-35,
2011.

[7] S. Houben, “A single target voting scheme for traffic sign detection”, in Proc.
IEEE IV Symp., pp. 124–129, June 2011.

[8] S. Xu, “Robust traffic sign shape recognition using geometric matching”,
IET Intell. Transp. Syst., vol. 3, no. 1, pp. 10–18, March 2009.

[9] F. Larsson and M. Felsberg, “Using Fourier descriptors and spatial models
for traffic sign recognition”, in Proc. Image Anal., pp. 238–249, 2011.

[10] D. Pei, F. Sun and H. Liu, “Supervised Low-Rank Matrix recovery for
Traffic Sign Recognition in Image Sequences”, IEEE Signal Processing
Letters, vol. 20, no. 3, March 2013.

[11] J. Stallkamp, M. Schlipsing, J. Salmen and C. Igel, “Man vs. computer:

Benchmarking machine learning algorithms for traffic sign recognition”, Neural
Networks, vol. 32, pp. 323-332, 2012.

[12] J. Park, J. Kwon, J. Oh, S. Lee, J-Y. Kim and H-j. Yoo, “A 92-mW Real-
Time Traffic Sign Recognition System with Robust Illumination Adaptation and
Support Vector Machine”, IEEE Journal of Solid-State Circuits, vol. 47, no. 11,
November 2012.

[13] J. Greenhalgh and M. Mirmehdi, “Real-Time Detection and Recognition of

Road Traffic Signs”, IEEE Transactions on Intelligent Transportation Systems,
vol. 13, no. 4, December 2012.

[14] C. Zi-xing and G. Ming-qin, “Traffic sign recognition algorithm based on

shape signature and dual-tree complex wavelet transform”, J. Cent. South
Univ., vol. 20, pp. 433-439, 2013.

45
[15] Zavadil, J.; Tuma, J.; Santos, V.M.F., “Traffic signs detection using blob
analysis and pattern recognition” Carpathian Control Conference (ICCC –
2012), pp. 776 – 779, 2012.

[16] Web pages of the OpenCV library, [online], 2014, Accessible on www:
http://opencv.org

[17] Web pages of the FANN library, [online], 2014, Accessible on www:
http://leenissen.dk/fann/wp/

[18] Web pages of IDS company, [online], 2014, Accessible on www:

http://en.ids-imaging.com/

[19] Online image dataset of traffic lights, [online], 2014, Accessible on:
http://lars.mec.ua.pt/public/LAR%20Projects/Vision/2011_JaromirZava
dil/Images/

[20] R. Hess, J.K. Moore and M.Hubbard, “Modeling the Manually

Controlled Bicycle,” IEEE Transactions on Systems, Man, and Cybernetics
- Part A: Systems and Humans , vol. 42, no.3, pp.545–
557, May 2012.

[21] Lukic, S.; Pantic, Z. Cutting the Cord: Static and Dynamic Inductive
Wireless Charging of Electric Vehicles. IEEE Electrif. Mag. 2013, 1, 57–64.

46