SlideShare a Scribd company logo
Garbage Collecting Robot Using YOLOv3 Deep
Learning Model
Akhila Assis
B.Tech Student
Department of ECE
RSET, Kochi
Kerala, India
Email: akhilaazziz1@gmail.com
Aleena Rose Biju
B.Tech Student
Department of ECE
RSET, Kochi
Kerala, India
Email: aleenasbiju@gmail.com
Alisha N A
B.Tech Student
Department of ECE
RSET, Kochi
Kerala, India
Email: alishanaushad1@gmail.com
Amrutha Dhanadas
B.Tech Student
Department of ECE
RSET, Kochi
Kerala, India
Email: amruthaanu99@gmail.com
Nitheesh Kurian
Assistant Professor
Department of ECE
RSET, Kochi
Kerala, India
Email: nitheeshk@rajagiritech.edu.in
Abstract—Due to rapid rise in population, there is an increase
in the waste scattered and littered on the roads. Waste manage-
ment has become an issue of concern, now more than ever around
the world. There are instances where waste is deliberately left
on the roads due to the lack of workers or means available. The
hazardous nature of the work and the health risks it can pose for
people is also another concern. With the rise of smart cities, we
recognized the need to develop an autonomous garbage collecting
robot that monitors, identifies and collects garbage from the
environment accurately. For this purpose, we also conducted a
comparative study of different object detection algorithms and
evaluated them based on various parameters.
Keywords—Garbage collecting robot, object detection al-
gorithm, CNN, YOLOv3, Autodesk TinkerCAD, Autodesk
Fusion 360, Transfer Learning
I. INTRODUCTION
One of the biggest problems that the world currently faces
is waste management. The streets are often littered with waste
consisting of bottles, can, food packets etc. The existing waste
management methods lack efficiency or incur high costs. The
need for an efficient and economical waste collection method
is the need of the hour.
This paper proposes a virtual idea of an autonomous
garbage collecting robot model that surveys an area, identifies
garbage lying on ground and collects them into the machine’s
mounted garbage collector. This can be considered a feasible
solution for proper waste management.
The robot is constructed with a four wheel drive on a
flat chassis. Four wheels are attached to the robot using four
different DC motors.Raspberry Pi is connected to an ultrasonic
sensor, and an Arduino UNO. The ultrasonic sensor detects
978-1-6654-3919-0/21/$31.00 ©2021 IEEE
any obstacle in front of it. If an object is detected at a distance
of 20 cm or less from the ultrasonic sensor, the robot stops
moving, the object detection flag is set as high, and the image
capturing module is activated. The image of the object is
captured with the help of a webcam. Based on this image,in
order to detect and distinguish garbage from other objects,
YOLOv3 (You Only Look Once) real-time object detection
algorithm is used.
The arm of the robot is designed as a grabber which
lift objects detected by robots as garbage.There are 2 servos
at the grabber : the first servo is to clamp the object and the
second one is to lift the grabber. The grabber is connected to
the Arduino UNO.
The virtual model of the robot is implemented using
Autodesk TinkerCAD and Autodesk Fusion 360.
Most of the existing work to make a garbage collecting
robot includes object detection using only sensor fusion,
computer vision algorithm, Iot etc.
In one method [1], an autonomous robot is proposed
to collect waste from an environment.The robot tested in a
designed region performs tasks such as avoiding the static and
dynamic obstacles and sucking of waste mechanism.In order
to clean the entire room efficiently, a combination of S-shaped
algorithm and wall follow algorithm is used
In another method [2] to detect and segregate into bio-
degradable and non-biodegradable, a robot is designed in a
manner that would be able to collect these items from the
environment when they were placed at random.To lift the
objects from the floor and place the objects in a container
an arm mechanism is used.
In order to make improvements over the existing methods
[3],a new technical idea of robotic arm is proposed in another
paper by designing a robotic arm with 7 DOF (Degree of
2021
10th
International
Conference
on
Advances
in
Computing
and
Communications
(ICACC)
|
978-1-6654-3919-0/21/$31.00
©2021
IEEE
|
DOI:
10.1109/ICACC-202152719.2021.9708298
Authorized licensed use limited to: Sharda University. Downloaded on May 24,2022 at 06:03:49 UTC from IEEE Xplore. Restrictions apply.
Freedom). The working starts at arm envelope, space and
increasing its efficiency.By using different algorithms and
simulation,the system is enabled to add grip,hold and place
capability. The traditional methods of designing a robotic arm
will be outperformed by the proposed approach.
In another paper [4] using a modified You Only Look
Once v3 (YOLOv3) method a vision-based water surface
garbage capture robot has been developed.The detection scales
of YOLOv3 are simplified from 3 to 2 to improve the real-time
detection performance.The anchor boxes of the training data
set are reclustered for replacing some of the original YOLOv3
prior anchor boxes that are not appropriate to the data set.
Detection speed and accuracy of the modified YOLOv3 were
found to be better than that of other object detection algo-
rithms.
II. DATASET
With more than 500 handpicked images, a dataset cus-
tomized for garbage collection robot was created. Five classes
of garbage were chosen namely bottle, can, food packet,
paper ball, and plastic bag. These images were then labelled
manually using LabelImg. LabelImg is a graphical annotation
tool which is used for labelling object bounding boxes in
images. After labelling an image classes.txt file is generated
which contains all the classes which is annotated. For every
image labelled there will a corresponding .txt file which
includes the metadata. The metadata will include object id,
center x, center y, width, height.
III. METHODOLOGY
The whole project can be described three phases:
• Object Detection
• Garbage Identification
• Garbage Collection
The block diagram and flow chart of the proposed model is
shown in figure 1 and figure 2 respectively.
Fig. 1. Block diagram
A. Object Detection
The garbage collecting robot is movable and it moves until
the ultrasonic sensor detects another object. Ultrasonic sensor
is capable of detecting objects of about 20 cm distance and
Fig. 2. Flow Chart
when an object is detected it will stop the dc motor so that
the robot doesn’t move further. Webcam will get activated and
the image will be send to the Raspberry pi. Garbage detection
using deep learning mechanism is started.
B. Garbage Detection
Garbage detection is done using YOLOv3 algorithm. The
captured image is given to Raspberry pi. There it ensures
that the correct object is detected as garbage. A comparative
study is done for YOLOv3 and Convolutional Neural Network
(CNN).
1) You Only Look Once (YOLO v3): A modified YOLOv3
network detects the garbage lying on the ground. YOLOv3
training was done using Google Colab. The object detection
code written in python was given to Raspberry pi 3b+ and run.
The garbage were detected.Transfer learning method is used
for YOLOv3 Object detection.
The overall prediction probability is calculated using the
formula:
Ntp
Ntp + Nfp
Ntp is the total number of correctly detected objects.
Nfp is the total number of incorrectly detected objects.
• Total number of correctly detected objects = 63
• Total number of incorrectly detected objects = 4
• Total number of object detected = 67
Using the above formula , the prediction probability of
YOLO v3 is calculated and 94.02 % efficiency was obtained.
Authorized licensed use limited to: Sharda University. Downloaded on May 24,2022 at 06:03:49 UTC from IEEE Xplore. Restrictions apply.
70 percent of images from the dataset was given for training
and 30 percent for testing.
2) Convolutional Neural Network (CNN) : Convolutional
Neural Network composes of various convolutional and pool-
ing layers. The input image is given to the first convolutional
layer and output is obtained as activation map. Filters are
applied to the convolutional layer which extract relevant
features from the input image to pass further. To minimize the
number of parameters, pooling layers are added. Before doing
the prediction, several convolutional and pooling layers are
added. The output layer in CNN is a fully connected layer. The
output is generated via output layer and a comparison is made
with error generation. This is then back propagated to update
the filter(weights) and bias values. For training and testing of
CNN, a dataset was created and the parameters were modified
for this project. The training was performed on Google Colab.
• Layer size is given as 64 and 3 convolutional layers were
used
• Filter used is 3x3 and the max pooling size is 2x2.
• In the dense layer, sigmoid activation function is used.
• Batch size = 32 and epoch = 100
• Validation split given as 70 percent for training and 30
percent for testing.
C. Garbage Collection
Using YOLOv3 algorithm, the Raspberry pi will transmit
a signal to the Arduino if the object detected is identified as
garbage. Arduino commence to operate the servo motors of
the garbage collecting robot. At first, the arm’s servo motor
causes the arms of the robot to go down and after that the
grabber servos rotate to collect the garbage. The arm’s servos
cause the arm to go upward and dispose the garbage to a bin
placed over the robot. If the object is not identified as garbage,
the robot continues to move forward.
IV. HARDWARE
A. Raspberry Pi 3B+
The Raspberry Pi is a mini computer board that comprises
of CPU, GPU, USB ports, I/O pins, WiFi, Bluetooth, USB
and network boot and power over ethernet facility. It is a
64 bit processor with 1GB RAM. It consists of Broadcom
BCM2837B0 chipset accompanied with with 1.4GHz Quad-
Core ARM Cortex-A53, 4 cores and consists of 40 pin header
(26 GPIOs). It has 4 USB, 2.0 ports Gigabit Ethernet and 2
pin set header. PoE (power over Ethernet) is a vital feature
included in the device.
B. Arduino Uno R3
Arduino Uno R3 is a microcontroller based on ATmega 328
that includes 14 digital I/O pins and 6 analog I/O pins.
• Operating Voltage : 5V
• Input voltage range : 7V to 12V
• Input voltage (limit) : 6V to 20 V
• DC Current for each I/O Pin : 20 mA
• DC Current used for 3.3V Pin : 50 mA.
• Flash Memory : -32 KB
• Speed of the CLK : 16 MHz
0.5 KB memory is used by the boot loader with SRAM 2 KB
and EEPROM 1 KB. It has an inbuilt LED.
C. Ultrasonic sensor
Ultrasonic sensor is a module which measures distance
using ultrasonic waves. The sensor head emits an ultrasonic
wave and receives the wave reflected back from the target.
It measures the distance to the target by measuring the time
between the emission and reception. To find the distance to
an object, the below mentioned formula can be used:
Distance = (speed of sound x time taken) / 2
• Supply voltage : 5V
• Global current consumption : 15 mA
• Ultrasonic Frequency : 40k Hz
• Maximal Range : 400 cm
• Minimal Range : 3 cm
• Resolution : 1 cm
• Trigger Pulse Width : 10 µs
• Outline Dimension : 43x20x15 mm
D. DC Motors
DC motor is a class of electrical machines that converts
direct current electrical power into mechanical power. The
most common types focus on the forces produced by magnetic
fields. In this project, two DC motors are used on the right
and the left each, giving a total of four.
E. Webcam
A webcam is a video camera that feeds or streams an image
or video in real time to or through a computer network,for
instance, the Internet. Webcam software helps users to record
a video or stream the video on the Internet. As video stream-
ing over the internet demands more bandwidth, compressed
formats are generally used.
F. Servomotors
A servomotor is a rotary actuator or linear actuator which
permits precise control of angular or linear position, velocity
and acceleration. It comprises of a suitable motor coupled to a
sensor for position feedback. This project uses 5 servo motors.
V. CIRCUIT SIMULATION OF ROBOT
The circuit simulation of the robot is done using a 3D
modelling online platform ’Autodesk Tinkercad’. The circuit
comprises of 5 servomotors and 4 DC motors. First servo
motor serves as the base of the structure followed by second
motor which is placed in the upper arm. Third and fourth
motor is placed in the lower arm and the final one serves as
gripper.
An ultrasonic distance sensor is used to calculate the
distance of the object to be detected. Threshold value of
distance is taken as 150 cm here. If an object is located at
a distance below 150 cm, four DC motors will start rotating
Authorized licensed use limited to: Sharda University. Downloaded on May 24,2022 at 06:03:49 UTC from IEEE Xplore. Restrictions apply.
simultaneously symbolizing the 4 wheels of the robot. When
an object is positioned at a distance beyond 150cm, rotation
of DC motor will stop and the servomotors will begin rotating
one after the other. This is to show the movement of the robotic
arm.
VI. AUTODESK FUSION 360
The three dimensional model of the garbage collecting robot
is done using Autodesk fusion 360. The model consists of
a base with four wheels and a robotic arm with grabber. It
also consist of five servo motors in which two servo motors
are placed in the lower arm and one each in gripper, upper
arm and the base. The servo motor used for this model is
Nema 23 servo motor. The Standard NEMA23 is 22.4Kgcm
holding torque. Rated for 200W AC servo applications up to
2000RPM.
VII. RESULT AND ANALYSIS
The comparison for different algorithms are listed in Table
1.
TABLE I
COMPARISON OF ALGORITHMS
Parameters YOLOv3 CNN
Accuracy 0.9341 0.849
Precision 0.9722 0.85
Recall 0.84 0.8491
F1 Score 0.9012 0.8512
The equation for Precision is given as:
TruePositives
TruePositives + FalsePositives
The equation for Recall is given as:
TruePositives
TruePositives + FalseNegatives
The equation for F1-score can thus be determined by:
2 ∗
Precision ∗ Recall
Precision + Recall
Based on these evaluation parameters, YOLOv3 is found to
have better detection than CNN. YOLOv3 has higher values
of precision, accuracy and F1 score but CNN has a greater
recall value.
TABLE II
CONFUSION MATRIX FOR YOLOV3 ALGORITHM
Classes Accuracy Precision Recall F1 Score
Plastic Bottle 0.9376 0.9131 0.8325 0.8709
Paper Ball 0.9149 0.9372 0.8662 0.9032
Can 0.9365 0.8922 0.8560 0.8737
Plastic bag 0.9642 0.9561 0.8753 0.9139
Food Packet 0.8567 0.8715 0.8393 0.8550
Fig. 3. Robotic circuit simulation
Fig. 4. 3D Model of the robot
Fig. 5. YOLOv3 Output
Authorized licensed use limited to: Sharda University. Downloaded on May 24,2022 at 06:03:49 UTC from IEEE Xplore. Restrictions apply.
Fig. 6. CNN Output
VIII. CONCLUSION
The proposed work was successfully implemented. Ob-
ject detection algorithms, YOLOv3 and CNN, were used
for garbage detection, for both image detection as well as
livestream. Transfer learning method was used for YOLOv3
algorithm in which dataset of five classes of garbage was cre-
ated and labelled using the LabellImg tool and training of the
dataset was done using Google colab. Finally testing was done
and got an accuracy of 0.9341%. For CNN algorithm same
set of dataset was used and acquired the accuracy of about
0.849%. These algorithms were evaluated based on different
parameters like Accuracy, Precision, Recall and F1 Score. A
virtual 3D model of the robot was created on Autodesk Fusion
360 platform. The circuit simulation was also done for the
robot in which the server motors and the DC motors move
based on object detection and object identification.
The current work only includes a virtual representation
of the robot. Therefore, the future goal involves bringing the
virtual model into a reality by making it. Garbage detection
can be evaluated using other image detection algorithms like
RCNN for better accuracy or speed. Also the number of
images in the dataset could be improved to increase accuracy.
The number of classes in the dataset can also be expanded
to improve the range of garbage that can be detected and col-
lected. Garbage segregation is also another possible expansion
of this project.
REFERENCES
[1] Xiali Li, Manjun Tian, Shihan Kong, Licheng Wu and Junzhi Yu,”A
modified YOLOv3 detection method for vision-based water surface
garbage capture robot”, International Journal of Advanced Robotic
Systems May-June 2020: 1–11
[2] Redmon J and Farhadi A. YOLOv3: an incremental improvement, arXiv
preprint arXiv:1804.02767, April 2020
[3] Daniel Octavian Melinte, Ana-Maria Travediu and Dan N. Du-
mitriu, ”Deep Convolutional Neural Networks Object Detector
for Real-Time Waste Identification”, Appl. Sci. 2020, 10, 7301;
doi:10.3390/app10207301
[4] Omkar Masurekar, Omkar Jadhav, Prateek Kulkarni, Shubham Patil,
”Real Time Object Detection Using YOLOv3”, International Research
Journal of Engineering and Technology (IRJET), Volume: 07 Issue: 03
Mar 2020.
[5] Richardson Santiago Teles de Menezes, Rafael Marrocos Magalhaes
and Helton Maia,” Object Recognition Using Convolutional Neural
Networks”, DOI: http://dx.doi.org/10.5772/intechopen.89726
[6] Virendra, Apoorva Mishra, Ritu Tiwari, Robotics Intelligent System
Design Lab ABV-Indian Institute of Information Technology Man-
agement, Gwalior, India, ”Robotic Gripper Arm System with Effective
Working Envelope”, International Conference on Intelligent Computing
and Control Systems (ICICCS 2018) IEEE Xplore Compliant Part
Number: CFP18K74-ART; ISBN:978-1-5386-2842-3
[7] Pranav Adarsh ,Pratibha , Manoj Kumar, ”YOLO v3-Tiny: Object
Detection and Recognition using one stage improved model”, 6th
International Conference on Advanced Computing & Communication
Systems (ICACCS) ,2020.
[8] Mr. Rakshith Ranganath, Ms. Bhawna Sharma, Ms. Pooja AR, Mr. Ro-
han C Jadhav, Ms.Asha A,”Autonomous Garbage Collecting Robot Wall-
E”,IJSRD - International Journal for Scientific Research Development,
ISSN (online): 2321-0613.
[9] G. Sivasankar,B. Durgalakshmi,K .Seyatha,”Autonomous Trash Collect-
ing Robot” International Journal of Engineering Research Technology
(IJERT) ISSN: 2278-0181.
[10] Saurav Kumar , Drishti Yadav , Himanshu Gupta , Om Prakash
Verma , Irshad Ahmad Ansari and Chang Wook Ahn , Department
of Instrumentation and Control Engineering, Dr. B. R. Ambedkar
National Institute of Technology Jalandhar, Punjab 144011, India ,
Novel YOLOv3 Algorithm-Based Deep Learning Approach for Waste
Segregation: Towards Smart Waste Management, Electronics 2021, 10,
14 https://dx.doi.org/10.3390/ electronics100100141.
Authorized licensed use limited to: Sharda University. Downloaded on May 24,2022 at 06:03:49 UTC from IEEE Xplore. Restrictions apply.

More Related Content

Similar to Garbage_Collecting_Robot_Using_YOLOv3_Deep_Learning_Model (1).pdf

PCL PROJECT PPT.pptx
PCL PROJECT PPT.pptxPCL PROJECT PPT.pptx
PCL PROJECT PPT.pptx
ShreyasGowda682614
 
Smart Bins : IOT Based Garbage Monitoring System
Smart Bins : IOT Based Garbage Monitoring SystemSmart Bins : IOT Based Garbage Monitoring System
Smart Bins : IOT Based Garbage Monitoring System
Kavish Khullar
 
SOCIAL DISTANCING DETECTION
SOCIAL DISTANCING DETECTIONSOCIAL DISTANCING DETECTION
SOCIAL DISTANCING DETECTION
IRJET Journal
 
SMART DUSTBINS FOR SMART CITIES
SMART DUSTBINS FOR SMART CITIESSMART DUSTBINS FOR SMART CITIES
SMART DUSTBINS FOR SMART CITIES
vivatechijri
 
A Robotic Prototype System for Child Monitoring
A Robotic Prototype System for Child MonitoringA Robotic Prototype System for Child Monitoring
A Robotic Prototype System for Child Monitoring
Waqas Tariq
 
IRJET - Swarm Robotic System for Mapping and Exploration of Extreme Envir...
IRJET -  	  Swarm Robotic System for Mapping and Exploration of Extreme Envir...IRJET -  	  Swarm Robotic System for Mapping and Exploration of Extreme Envir...
IRJET - Swarm Robotic System for Mapping and Exploration of Extreme Envir...
IRJET Journal
 
Road signs detection using voila jone's algorithm with the help of opencv
Road signs detection using voila jone's algorithm with the help of opencvRoad signs detection using voila jone's algorithm with the help of opencv
Road signs detection using voila jone's algorithm with the help of opencv
MohdSalim34
 
An assistive model of obstacle detection based on deep learning: YOLOv3 for v...
An assistive model of obstacle detection based on deep learning: YOLOv3 for v...An assistive model of obstacle detection based on deep learning: YOLOv3 for v...
An assistive model of obstacle detection based on deep learning: YOLOv3 for v...
IJECEIAES
 
IISc Internship Report
IISc Internship ReportIISc Internship Report
IISc Internship Report
HarshilJain26
 
Color Tracking Robot
Color Tracking RobotColor Tracking Robot
Color Tracking Robot
paperpublications3
 
www.ijerd.com
www.ijerd.comwww.ijerd.com
www.ijerd.com
IJERD Editor
 
HCI for Real world Applications
HCI for Real world ApplicationsHCI for Real world Applications
HCI for Real world Applications
IOSR Journals
 
L01117074
L01117074L01117074
L01117074
IOSR Journals
 
Landmines Detection by Robots presentation
Landmines Detection by Robots  presentationLandmines Detection by Robots  presentation
Landmines Detection by Robots presentation
Ahmed Abdelaziz
 
IRJET-Cleaner Drone
IRJET-Cleaner DroneIRJET-Cleaner Drone
IRJET-Cleaner Drone
IRJET Journal
 
Object Detection An Overview
Object Detection An OverviewObject Detection An Overview
Object Detection An Overview
ijtsrd
 
IRJET- Real-Time Object Detection using Deep Learning: A Survey
IRJET- Real-Time Object Detection using Deep Learning: A SurveyIRJET- Real-Time Object Detection using Deep Learning: A Survey
IRJET- Real-Time Object Detection using Deep Learning: A Survey
IRJET Journal
 
Simulation of collision avoidance by navigation
Simulation of collision avoidance by navigationSimulation of collision avoidance by navigation
Simulation of collision avoidance by navigation
eSAT Publishing House
 
IRJET- Determining the Components of Waste Assisted with Analysis of Meth...
IRJET-  	  Determining the Components of Waste Assisted with Analysis of Meth...IRJET-  	  Determining the Components of Waste Assisted with Analysis of Meth...
IRJET- Determining the Components of Waste Assisted with Analysis of Meth...
IRJET Journal
 
Virtual environment for assistant mobile robot
Virtual environment for assistant mobile robotVirtual environment for assistant mobile robot
Virtual environment for assistant mobile robot
IJECEIAES
 

Similar to Garbage_Collecting_Robot_Using_YOLOv3_Deep_Learning_Model (1).pdf (20)

PCL PROJECT PPT.pptx
PCL PROJECT PPT.pptxPCL PROJECT PPT.pptx
PCL PROJECT PPT.pptx
 
Smart Bins : IOT Based Garbage Monitoring System
Smart Bins : IOT Based Garbage Monitoring SystemSmart Bins : IOT Based Garbage Monitoring System
Smart Bins : IOT Based Garbage Monitoring System
 
SOCIAL DISTANCING DETECTION
SOCIAL DISTANCING DETECTIONSOCIAL DISTANCING DETECTION
SOCIAL DISTANCING DETECTION
 
SMART DUSTBINS FOR SMART CITIES
SMART DUSTBINS FOR SMART CITIESSMART DUSTBINS FOR SMART CITIES
SMART DUSTBINS FOR SMART CITIES
 
A Robotic Prototype System for Child Monitoring
A Robotic Prototype System for Child MonitoringA Robotic Prototype System for Child Monitoring
A Robotic Prototype System for Child Monitoring
 
IRJET - Swarm Robotic System for Mapping and Exploration of Extreme Envir...
IRJET -  	  Swarm Robotic System for Mapping and Exploration of Extreme Envir...IRJET -  	  Swarm Robotic System for Mapping and Exploration of Extreme Envir...
IRJET - Swarm Robotic System for Mapping and Exploration of Extreme Envir...
 
Road signs detection using voila jone's algorithm with the help of opencv
Road signs detection using voila jone's algorithm with the help of opencvRoad signs detection using voila jone's algorithm with the help of opencv
Road signs detection using voila jone's algorithm with the help of opencv
 
An assistive model of obstacle detection based on deep learning: YOLOv3 for v...
An assistive model of obstacle detection based on deep learning: YOLOv3 for v...An assistive model of obstacle detection based on deep learning: YOLOv3 for v...
An assistive model of obstacle detection based on deep learning: YOLOv3 for v...
 
IISc Internship Report
IISc Internship ReportIISc Internship Report
IISc Internship Report
 
Color Tracking Robot
Color Tracking RobotColor Tracking Robot
Color Tracking Robot
 
www.ijerd.com
www.ijerd.comwww.ijerd.com
www.ijerd.com
 
HCI for Real world Applications
HCI for Real world ApplicationsHCI for Real world Applications
HCI for Real world Applications
 
L01117074
L01117074L01117074
L01117074
 
Landmines Detection by Robots presentation
Landmines Detection by Robots  presentationLandmines Detection by Robots  presentation
Landmines Detection by Robots presentation
 
IRJET-Cleaner Drone
IRJET-Cleaner DroneIRJET-Cleaner Drone
IRJET-Cleaner Drone
 
Object Detection An Overview
Object Detection An OverviewObject Detection An Overview
Object Detection An Overview
 
IRJET- Real-Time Object Detection using Deep Learning: A Survey
IRJET- Real-Time Object Detection using Deep Learning: A SurveyIRJET- Real-Time Object Detection using Deep Learning: A Survey
IRJET- Real-Time Object Detection using Deep Learning: A Survey
 
Simulation of collision avoidance by navigation
Simulation of collision avoidance by navigationSimulation of collision avoidance by navigation
Simulation of collision avoidance by navigation
 
IRJET- Determining the Components of Waste Assisted with Analysis of Meth...
IRJET-  	  Determining the Components of Waste Assisted with Analysis of Meth...IRJET-  	  Determining the Components of Waste Assisted with Analysis of Meth...
IRJET- Determining the Components of Waste Assisted with Analysis of Meth...
 
Virtual environment for assistant mobile robot
Virtual environment for assistant mobile robotVirtual environment for assistant mobile robot
Virtual environment for assistant mobile robot
 

Recently uploaded

AHMED TALAAT ARCHITECTURE PORTFOLIO .pdf
AHMED TALAAT ARCHITECTURE PORTFOLIO .pdfAHMED TALAAT ARCHITECTURE PORTFOLIO .pdf
AHMED TALAAT ARCHITECTURE PORTFOLIO .pdf
talaatahm
 
一比一原版马来西亚世纪大学毕业证成绩单一模一样
一比一原版马来西亚世纪大学毕业证成绩单一模一样一比一原版马来西亚世纪大学毕业证成绩单一模一样
一比一原版马来西亚世纪大学毕业证成绩单一模一样
k4krdgxx
 
NHL Stenden University of Applied Sciences Diploma Degree Transcript
NHL Stenden University of Applied Sciences Diploma Degree TranscriptNHL Stenden University of Applied Sciences Diploma Degree Transcript
NHL Stenden University of Applied Sciences Diploma Degree Transcript
lhtvqoag
 
Getting Data Ready for Culture Hack by Neontribe
Getting Data Ready for Culture Hack by NeontribeGetting Data Ready for Culture Hack by Neontribe
Getting Data Ready for Culture Hack by Neontribe
Harry Harrold
 
Introduction to User experience design for beginner
Introduction to User experience design for beginnerIntroduction to User experience design for beginner
Introduction to User experience design for beginner
ellemjani
 
UXpert_Report (UALR Mapping Renewal 2022).pdf
UXpert_Report (UALR Mapping Renewal 2022).pdfUXpert_Report (UALR Mapping Renewal 2022).pdf
UXpert_Report (UALR Mapping Renewal 2022).pdf
anthonylin333
 
Heuristics Evaluation - How to Guide.pdf
Heuristics Evaluation - How to Guide.pdfHeuristics Evaluation - How to Guide.pdf
Heuristics Evaluation - How to Guide.pdf
Jaime Brown
 
一比一原版马里兰大学毕业证(UMD毕业证书)如何办理
一比一原版马里兰大学毕业证(UMD毕业证书)如何办理一比一原版马里兰大学毕业证(UMD毕业证书)如何办理
一比一原版马里兰大学毕业证(UMD毕业证书)如何办理
9lq7ultg
 
一比一原版(Vancouver毕业证书)温哥华岛大学毕业证如何办理
一比一原版(Vancouver毕业证书)温哥华岛大学毕业证如何办理一比一原版(Vancouver毕业证书)温哥华岛大学毕业证如何办理
一比一原版(Vancouver毕业证书)温哥华岛大学毕业证如何办理
ijk38lw
 
一比一原版(UoN毕业证书)纽卡斯尔大学毕业证如何办理
一比一原版(UoN毕业证书)纽卡斯尔大学毕业证如何办理一比一原版(UoN毕业证书)纽卡斯尔大学毕业证如何办理
一比一原版(UoN毕业证书)纽卡斯尔大学毕业证如何办理
f22b6g9c
 
原版制作(MDIS毕业证书)新加坡管理发展学院毕业证学位证一模一样
原版制作(MDIS毕业证书)新加坡管理发展学院毕业证学位证一模一样原版制作(MDIS毕业证书)新加坡管理发展学院毕业证学位证一模一样
原版制作(MDIS毕业证书)新加坡管理发展学院毕业证学位证一模一样
hw2xf1m
 
一比一原版(CSU毕业证书)查尔斯特大学毕业证如何办理
一比一原版(CSU毕业证书)查尔斯特大学毕业证如何办理一比一原版(CSU毕业证书)查尔斯特大学毕业证如何办理
一比一原版(CSU毕业证书)查尔斯特大学毕业证如何办理
67n7f53
 
哪里办理美国中央华盛顿大学毕业证双学位证书原版一模一样
哪里办理美国中央华盛顿大学毕业证双学位证书原版一模一样哪里办理美国中央华盛顿大学毕业证双学位证书原版一模一样
哪里办理美国中央华盛顿大学毕业证双学位证书原版一模一样
qo1as76n
 
Practical eLearning Makeovers for Everyone
Practical eLearning Makeovers for EveryonePractical eLearning Makeovers for Everyone
Practical eLearning Makeovers for Everyone
Bianca Woods
 
一比一原版美国哥伦比亚大学毕业证Columbia成绩单一模一样
一比一原版美国哥伦比亚大学毕业证Columbia成绩单一模一样一比一原版美国哥伦比亚大学毕业证Columbia成绩单一模一样
一比一原版美国哥伦比亚大学毕业证Columbia成绩单一模一样
881evgn0
 
按照学校原版(UIUC文凭证书)伊利诺伊大学|厄巴纳-香槟分校毕业证快速办理
按照学校原版(UIUC文凭证书)伊利诺伊大学|厄巴纳-香槟分校毕业证快速办理按照学校原版(UIUC文凭证书)伊利诺伊大学|厄巴纳-香槟分校毕业证快速办理
按照学校原版(UIUC文凭证书)伊利诺伊大学|厄巴纳-香槟分校毕业证快速办理
kuapy
 
modular-kitchen home plan civil engineering.pdf
modular-kitchen home plan civil engineering.pdfmodular-kitchen home plan civil engineering.pdf
modular-kitchen home plan civil engineering.pdf
RashmitaSwain3
 
Manual ISH (International Society of Hypertension)
Manual ISH (International Society of Hypertension)Manual ISH (International Society of Hypertension)
Manual ISH (International Society of Hypertension)
bagmai
 
一比一原版亚利桑那大学毕业证(UA毕业证书)如何办理
一比一原版亚利桑那大学毕业证(UA毕业证书)如何办理一比一原版亚利桑那大学毕业证(UA毕业证书)如何办理
一比一原版亚利桑那大学毕业证(UA毕业证书)如何办理
21uul8se
 
一比一原版(LaTrobe毕业证书)拉筹伯大学毕业证如何办理
一比一原版(LaTrobe毕业证书)拉筹伯大学毕业证如何办理一比一原版(LaTrobe毕业证书)拉筹伯大学毕业证如何办理
一比一原版(LaTrobe毕业证书)拉筹伯大学毕业证如何办理
67n7f53
 

Recently uploaded (20)

AHMED TALAAT ARCHITECTURE PORTFOLIO .pdf
AHMED TALAAT ARCHITECTURE PORTFOLIO .pdfAHMED TALAAT ARCHITECTURE PORTFOLIO .pdf
AHMED TALAAT ARCHITECTURE PORTFOLIO .pdf
 
一比一原版马来西亚世纪大学毕业证成绩单一模一样
一比一原版马来西亚世纪大学毕业证成绩单一模一样一比一原版马来西亚世纪大学毕业证成绩单一模一样
一比一原版马来西亚世纪大学毕业证成绩单一模一样
 
NHL Stenden University of Applied Sciences Diploma Degree Transcript
NHL Stenden University of Applied Sciences Diploma Degree TranscriptNHL Stenden University of Applied Sciences Diploma Degree Transcript
NHL Stenden University of Applied Sciences Diploma Degree Transcript
 
Getting Data Ready for Culture Hack by Neontribe
Getting Data Ready for Culture Hack by NeontribeGetting Data Ready for Culture Hack by Neontribe
Getting Data Ready for Culture Hack by Neontribe
 
Introduction to User experience design for beginner
Introduction to User experience design for beginnerIntroduction to User experience design for beginner
Introduction to User experience design for beginner
 
UXpert_Report (UALR Mapping Renewal 2022).pdf
UXpert_Report (UALR Mapping Renewal 2022).pdfUXpert_Report (UALR Mapping Renewal 2022).pdf
UXpert_Report (UALR Mapping Renewal 2022).pdf
 
Heuristics Evaluation - How to Guide.pdf
Heuristics Evaluation - How to Guide.pdfHeuristics Evaluation - How to Guide.pdf
Heuristics Evaluation - How to Guide.pdf
 
一比一原版马里兰大学毕业证(UMD毕业证书)如何办理
一比一原版马里兰大学毕业证(UMD毕业证书)如何办理一比一原版马里兰大学毕业证(UMD毕业证书)如何办理
一比一原版马里兰大学毕业证(UMD毕业证书)如何办理
 
一比一原版(Vancouver毕业证书)温哥华岛大学毕业证如何办理
一比一原版(Vancouver毕业证书)温哥华岛大学毕业证如何办理一比一原版(Vancouver毕业证书)温哥华岛大学毕业证如何办理
一比一原版(Vancouver毕业证书)温哥华岛大学毕业证如何办理
 
一比一原版(UoN毕业证书)纽卡斯尔大学毕业证如何办理
一比一原版(UoN毕业证书)纽卡斯尔大学毕业证如何办理一比一原版(UoN毕业证书)纽卡斯尔大学毕业证如何办理
一比一原版(UoN毕业证书)纽卡斯尔大学毕业证如何办理
 
原版制作(MDIS毕业证书)新加坡管理发展学院毕业证学位证一模一样
原版制作(MDIS毕业证书)新加坡管理发展学院毕业证学位证一模一样原版制作(MDIS毕业证书)新加坡管理发展学院毕业证学位证一模一样
原版制作(MDIS毕业证书)新加坡管理发展学院毕业证学位证一模一样
 
一比一原版(CSU毕业证书)查尔斯特大学毕业证如何办理
一比一原版(CSU毕业证书)查尔斯特大学毕业证如何办理一比一原版(CSU毕业证书)查尔斯特大学毕业证如何办理
一比一原版(CSU毕业证书)查尔斯特大学毕业证如何办理
 
哪里办理美国中央华盛顿大学毕业证双学位证书原版一模一样
哪里办理美国中央华盛顿大学毕业证双学位证书原版一模一样哪里办理美国中央华盛顿大学毕业证双学位证书原版一模一样
哪里办理美国中央华盛顿大学毕业证双学位证书原版一模一样
 
Practical eLearning Makeovers for Everyone
Practical eLearning Makeovers for EveryonePractical eLearning Makeovers for Everyone
Practical eLearning Makeovers for Everyone
 
一比一原版美国哥伦比亚大学毕业证Columbia成绩单一模一样
一比一原版美国哥伦比亚大学毕业证Columbia成绩单一模一样一比一原版美国哥伦比亚大学毕业证Columbia成绩单一模一样
一比一原版美国哥伦比亚大学毕业证Columbia成绩单一模一样
 
按照学校原版(UIUC文凭证书)伊利诺伊大学|厄巴纳-香槟分校毕业证快速办理
按照学校原版(UIUC文凭证书)伊利诺伊大学|厄巴纳-香槟分校毕业证快速办理按照学校原版(UIUC文凭证书)伊利诺伊大学|厄巴纳-香槟分校毕业证快速办理
按照学校原版(UIUC文凭证书)伊利诺伊大学|厄巴纳-香槟分校毕业证快速办理
 
modular-kitchen home plan civil engineering.pdf
modular-kitchen home plan civil engineering.pdfmodular-kitchen home plan civil engineering.pdf
modular-kitchen home plan civil engineering.pdf
 
Manual ISH (International Society of Hypertension)
Manual ISH (International Society of Hypertension)Manual ISH (International Society of Hypertension)
Manual ISH (International Society of Hypertension)
 
一比一原版亚利桑那大学毕业证(UA毕业证书)如何办理
一比一原版亚利桑那大学毕业证(UA毕业证书)如何办理一比一原版亚利桑那大学毕业证(UA毕业证书)如何办理
一比一原版亚利桑那大学毕业证(UA毕业证书)如何办理
 
一比一原版(LaTrobe毕业证书)拉筹伯大学毕业证如何办理
一比一原版(LaTrobe毕业证书)拉筹伯大学毕业证如何办理一比一原版(LaTrobe毕业证书)拉筹伯大学毕业证如何办理
一比一原版(LaTrobe毕业证书)拉筹伯大学毕业证如何办理
 

Garbage_Collecting_Robot_Using_YOLOv3_Deep_Learning_Model (1).pdf

  • 1. Garbage Collecting Robot Using YOLOv3 Deep Learning Model Akhila Assis B.Tech Student Department of ECE RSET, Kochi Kerala, India Email: akhilaazziz1@gmail.com Aleena Rose Biju B.Tech Student Department of ECE RSET, Kochi Kerala, India Email: aleenasbiju@gmail.com Alisha N A B.Tech Student Department of ECE RSET, Kochi Kerala, India Email: alishanaushad1@gmail.com Amrutha Dhanadas B.Tech Student Department of ECE RSET, Kochi Kerala, India Email: amruthaanu99@gmail.com Nitheesh Kurian Assistant Professor Department of ECE RSET, Kochi Kerala, India Email: nitheeshk@rajagiritech.edu.in Abstract—Due to rapid rise in population, there is an increase in the waste scattered and littered on the roads. Waste manage- ment has become an issue of concern, now more than ever around the world. There are instances where waste is deliberately left on the roads due to the lack of workers or means available. The hazardous nature of the work and the health risks it can pose for people is also another concern. With the rise of smart cities, we recognized the need to develop an autonomous garbage collecting robot that monitors, identifies and collects garbage from the environment accurately. For this purpose, we also conducted a comparative study of different object detection algorithms and evaluated them based on various parameters. Keywords—Garbage collecting robot, object detection al- gorithm, CNN, YOLOv3, Autodesk TinkerCAD, Autodesk Fusion 360, Transfer Learning I. INTRODUCTION One of the biggest problems that the world currently faces is waste management. The streets are often littered with waste consisting of bottles, can, food packets etc. The existing waste management methods lack efficiency or incur high costs. The need for an efficient and economical waste collection method is the need of the hour. This paper proposes a virtual idea of an autonomous garbage collecting robot model that surveys an area, identifies garbage lying on ground and collects them into the machine’s mounted garbage collector. This can be considered a feasible solution for proper waste management. The robot is constructed with a four wheel drive on a flat chassis. Four wheels are attached to the robot using four different DC motors.Raspberry Pi is connected to an ultrasonic sensor, and an Arduino UNO. The ultrasonic sensor detects 978-1-6654-3919-0/21/$31.00 ©2021 IEEE any obstacle in front of it. If an object is detected at a distance of 20 cm or less from the ultrasonic sensor, the robot stops moving, the object detection flag is set as high, and the image capturing module is activated. The image of the object is captured with the help of a webcam. Based on this image,in order to detect and distinguish garbage from other objects, YOLOv3 (You Only Look Once) real-time object detection algorithm is used. The arm of the robot is designed as a grabber which lift objects detected by robots as garbage.There are 2 servos at the grabber : the first servo is to clamp the object and the second one is to lift the grabber. The grabber is connected to the Arduino UNO. The virtual model of the robot is implemented using Autodesk TinkerCAD and Autodesk Fusion 360. Most of the existing work to make a garbage collecting robot includes object detection using only sensor fusion, computer vision algorithm, Iot etc. In one method [1], an autonomous robot is proposed to collect waste from an environment.The robot tested in a designed region performs tasks such as avoiding the static and dynamic obstacles and sucking of waste mechanism.In order to clean the entire room efficiently, a combination of S-shaped algorithm and wall follow algorithm is used In another method [2] to detect and segregate into bio- degradable and non-biodegradable, a robot is designed in a manner that would be able to collect these items from the environment when they were placed at random.To lift the objects from the floor and place the objects in a container an arm mechanism is used. In order to make improvements over the existing methods [3],a new technical idea of robotic arm is proposed in another paper by designing a robotic arm with 7 DOF (Degree of 2021 10th International Conference on Advances in Computing and Communications (ICACC) | 978-1-6654-3919-0/21/$31.00 ©2021 IEEE | DOI: 10.1109/ICACC-202152719.2021.9708298 Authorized licensed use limited to: Sharda University. Downloaded on May 24,2022 at 06:03:49 UTC from IEEE Xplore. Restrictions apply.
  • 2. Freedom). The working starts at arm envelope, space and increasing its efficiency.By using different algorithms and simulation,the system is enabled to add grip,hold and place capability. The traditional methods of designing a robotic arm will be outperformed by the proposed approach. In another paper [4] using a modified You Only Look Once v3 (YOLOv3) method a vision-based water surface garbage capture robot has been developed.The detection scales of YOLOv3 are simplified from 3 to 2 to improve the real-time detection performance.The anchor boxes of the training data set are reclustered for replacing some of the original YOLOv3 prior anchor boxes that are not appropriate to the data set. Detection speed and accuracy of the modified YOLOv3 were found to be better than that of other object detection algo- rithms. II. DATASET With more than 500 handpicked images, a dataset cus- tomized for garbage collection robot was created. Five classes of garbage were chosen namely bottle, can, food packet, paper ball, and plastic bag. These images were then labelled manually using LabelImg. LabelImg is a graphical annotation tool which is used for labelling object bounding boxes in images. After labelling an image classes.txt file is generated which contains all the classes which is annotated. For every image labelled there will a corresponding .txt file which includes the metadata. The metadata will include object id, center x, center y, width, height. III. METHODOLOGY The whole project can be described three phases: • Object Detection • Garbage Identification • Garbage Collection The block diagram and flow chart of the proposed model is shown in figure 1 and figure 2 respectively. Fig. 1. Block diagram A. Object Detection The garbage collecting robot is movable and it moves until the ultrasonic sensor detects another object. Ultrasonic sensor is capable of detecting objects of about 20 cm distance and Fig. 2. Flow Chart when an object is detected it will stop the dc motor so that the robot doesn’t move further. Webcam will get activated and the image will be send to the Raspberry pi. Garbage detection using deep learning mechanism is started. B. Garbage Detection Garbage detection is done using YOLOv3 algorithm. The captured image is given to Raspberry pi. There it ensures that the correct object is detected as garbage. A comparative study is done for YOLOv3 and Convolutional Neural Network (CNN). 1) You Only Look Once (YOLO v3): A modified YOLOv3 network detects the garbage lying on the ground. YOLOv3 training was done using Google Colab. The object detection code written in python was given to Raspberry pi 3b+ and run. The garbage were detected.Transfer learning method is used for YOLOv3 Object detection. The overall prediction probability is calculated using the formula: Ntp Ntp + Nfp Ntp is the total number of correctly detected objects. Nfp is the total number of incorrectly detected objects. • Total number of correctly detected objects = 63 • Total number of incorrectly detected objects = 4 • Total number of object detected = 67 Using the above formula , the prediction probability of YOLO v3 is calculated and 94.02 % efficiency was obtained. Authorized licensed use limited to: Sharda University. Downloaded on May 24,2022 at 06:03:49 UTC from IEEE Xplore. Restrictions apply.
  • 3. 70 percent of images from the dataset was given for training and 30 percent for testing. 2) Convolutional Neural Network (CNN) : Convolutional Neural Network composes of various convolutional and pool- ing layers. The input image is given to the first convolutional layer and output is obtained as activation map. Filters are applied to the convolutional layer which extract relevant features from the input image to pass further. To minimize the number of parameters, pooling layers are added. Before doing the prediction, several convolutional and pooling layers are added. The output layer in CNN is a fully connected layer. The output is generated via output layer and a comparison is made with error generation. This is then back propagated to update the filter(weights) and bias values. For training and testing of CNN, a dataset was created and the parameters were modified for this project. The training was performed on Google Colab. • Layer size is given as 64 and 3 convolutional layers were used • Filter used is 3x3 and the max pooling size is 2x2. • In the dense layer, sigmoid activation function is used. • Batch size = 32 and epoch = 100 • Validation split given as 70 percent for training and 30 percent for testing. C. Garbage Collection Using YOLOv3 algorithm, the Raspberry pi will transmit a signal to the Arduino if the object detected is identified as garbage. Arduino commence to operate the servo motors of the garbage collecting robot. At first, the arm’s servo motor causes the arms of the robot to go down and after that the grabber servos rotate to collect the garbage. The arm’s servos cause the arm to go upward and dispose the garbage to a bin placed over the robot. If the object is not identified as garbage, the robot continues to move forward. IV. HARDWARE A. Raspberry Pi 3B+ The Raspberry Pi is a mini computer board that comprises of CPU, GPU, USB ports, I/O pins, WiFi, Bluetooth, USB and network boot and power over ethernet facility. It is a 64 bit processor with 1GB RAM. It consists of Broadcom BCM2837B0 chipset accompanied with with 1.4GHz Quad- Core ARM Cortex-A53, 4 cores and consists of 40 pin header (26 GPIOs). It has 4 USB, 2.0 ports Gigabit Ethernet and 2 pin set header. PoE (power over Ethernet) is a vital feature included in the device. B. Arduino Uno R3 Arduino Uno R3 is a microcontroller based on ATmega 328 that includes 14 digital I/O pins and 6 analog I/O pins. • Operating Voltage : 5V • Input voltage range : 7V to 12V • Input voltage (limit) : 6V to 20 V • DC Current for each I/O Pin : 20 mA • DC Current used for 3.3V Pin : 50 mA. • Flash Memory : -32 KB • Speed of the CLK : 16 MHz 0.5 KB memory is used by the boot loader with SRAM 2 KB and EEPROM 1 KB. It has an inbuilt LED. C. Ultrasonic sensor Ultrasonic sensor is a module which measures distance using ultrasonic waves. The sensor head emits an ultrasonic wave and receives the wave reflected back from the target. It measures the distance to the target by measuring the time between the emission and reception. To find the distance to an object, the below mentioned formula can be used: Distance = (speed of sound x time taken) / 2 • Supply voltage : 5V • Global current consumption : 15 mA • Ultrasonic Frequency : 40k Hz • Maximal Range : 400 cm • Minimal Range : 3 cm • Resolution : 1 cm • Trigger Pulse Width : 10 µs • Outline Dimension : 43x20x15 mm D. DC Motors DC motor is a class of electrical machines that converts direct current electrical power into mechanical power. The most common types focus on the forces produced by magnetic fields. In this project, two DC motors are used on the right and the left each, giving a total of four. E. Webcam A webcam is a video camera that feeds or streams an image or video in real time to or through a computer network,for instance, the Internet. Webcam software helps users to record a video or stream the video on the Internet. As video stream- ing over the internet demands more bandwidth, compressed formats are generally used. F. Servomotors A servomotor is a rotary actuator or linear actuator which permits precise control of angular or linear position, velocity and acceleration. It comprises of a suitable motor coupled to a sensor for position feedback. This project uses 5 servo motors. V. CIRCUIT SIMULATION OF ROBOT The circuit simulation of the robot is done using a 3D modelling online platform ’Autodesk Tinkercad’. The circuit comprises of 5 servomotors and 4 DC motors. First servo motor serves as the base of the structure followed by second motor which is placed in the upper arm. Third and fourth motor is placed in the lower arm and the final one serves as gripper. An ultrasonic distance sensor is used to calculate the distance of the object to be detected. Threshold value of distance is taken as 150 cm here. If an object is located at a distance below 150 cm, four DC motors will start rotating Authorized licensed use limited to: Sharda University. Downloaded on May 24,2022 at 06:03:49 UTC from IEEE Xplore. Restrictions apply.
  • 4. simultaneously symbolizing the 4 wheels of the robot. When an object is positioned at a distance beyond 150cm, rotation of DC motor will stop and the servomotors will begin rotating one after the other. This is to show the movement of the robotic arm. VI. AUTODESK FUSION 360 The three dimensional model of the garbage collecting robot is done using Autodesk fusion 360. The model consists of a base with four wheels and a robotic arm with grabber. It also consist of five servo motors in which two servo motors are placed in the lower arm and one each in gripper, upper arm and the base. The servo motor used for this model is Nema 23 servo motor. The Standard NEMA23 is 22.4Kgcm holding torque. Rated for 200W AC servo applications up to 2000RPM. VII. RESULT AND ANALYSIS The comparison for different algorithms are listed in Table 1. TABLE I COMPARISON OF ALGORITHMS Parameters YOLOv3 CNN Accuracy 0.9341 0.849 Precision 0.9722 0.85 Recall 0.84 0.8491 F1 Score 0.9012 0.8512 The equation for Precision is given as: TruePositives TruePositives + FalsePositives The equation for Recall is given as: TruePositives TruePositives + FalseNegatives The equation for F1-score can thus be determined by: 2 ∗ Precision ∗ Recall Precision + Recall Based on these evaluation parameters, YOLOv3 is found to have better detection than CNN. YOLOv3 has higher values of precision, accuracy and F1 score but CNN has a greater recall value. TABLE II CONFUSION MATRIX FOR YOLOV3 ALGORITHM Classes Accuracy Precision Recall F1 Score Plastic Bottle 0.9376 0.9131 0.8325 0.8709 Paper Ball 0.9149 0.9372 0.8662 0.9032 Can 0.9365 0.8922 0.8560 0.8737 Plastic bag 0.9642 0.9561 0.8753 0.9139 Food Packet 0.8567 0.8715 0.8393 0.8550 Fig. 3. Robotic circuit simulation Fig. 4. 3D Model of the robot Fig. 5. YOLOv3 Output Authorized licensed use limited to: Sharda University. Downloaded on May 24,2022 at 06:03:49 UTC from IEEE Xplore. Restrictions apply.
  • 5. Fig. 6. CNN Output VIII. CONCLUSION The proposed work was successfully implemented. Ob- ject detection algorithms, YOLOv3 and CNN, were used for garbage detection, for both image detection as well as livestream. Transfer learning method was used for YOLOv3 algorithm in which dataset of five classes of garbage was cre- ated and labelled using the LabellImg tool and training of the dataset was done using Google colab. Finally testing was done and got an accuracy of 0.9341%. For CNN algorithm same set of dataset was used and acquired the accuracy of about 0.849%. These algorithms were evaluated based on different parameters like Accuracy, Precision, Recall and F1 Score. A virtual 3D model of the robot was created on Autodesk Fusion 360 platform. The circuit simulation was also done for the robot in which the server motors and the DC motors move based on object detection and object identification. The current work only includes a virtual representation of the robot. Therefore, the future goal involves bringing the virtual model into a reality by making it. Garbage detection can be evaluated using other image detection algorithms like RCNN for better accuracy or speed. Also the number of images in the dataset could be improved to increase accuracy. The number of classes in the dataset can also be expanded to improve the range of garbage that can be detected and col- lected. Garbage segregation is also another possible expansion of this project. REFERENCES [1] Xiali Li, Manjun Tian, Shihan Kong, Licheng Wu and Junzhi Yu,”A modified YOLOv3 detection method for vision-based water surface garbage capture robot”, International Journal of Advanced Robotic Systems May-June 2020: 1–11 [2] Redmon J and Farhadi A. YOLOv3: an incremental improvement, arXiv preprint arXiv:1804.02767, April 2020 [3] Daniel Octavian Melinte, Ana-Maria Travediu and Dan N. Du- mitriu, ”Deep Convolutional Neural Networks Object Detector for Real-Time Waste Identification”, Appl. Sci. 2020, 10, 7301; doi:10.3390/app10207301 [4] Omkar Masurekar, Omkar Jadhav, Prateek Kulkarni, Shubham Patil, ”Real Time Object Detection Using YOLOv3”, International Research Journal of Engineering and Technology (IRJET), Volume: 07 Issue: 03 Mar 2020. [5] Richardson Santiago Teles de Menezes, Rafael Marrocos Magalhaes and Helton Maia,” Object Recognition Using Convolutional Neural Networks”, DOI: http://dx.doi.org/10.5772/intechopen.89726 [6] Virendra, Apoorva Mishra, Ritu Tiwari, Robotics Intelligent System Design Lab ABV-Indian Institute of Information Technology Man- agement, Gwalior, India, ”Robotic Gripper Arm System with Effective Working Envelope”, International Conference on Intelligent Computing and Control Systems (ICICCS 2018) IEEE Xplore Compliant Part Number: CFP18K74-ART; ISBN:978-1-5386-2842-3 [7] Pranav Adarsh ,Pratibha , Manoj Kumar, ”YOLO v3-Tiny: Object Detection and Recognition using one stage improved model”, 6th International Conference on Advanced Computing & Communication Systems (ICACCS) ,2020. [8] Mr. Rakshith Ranganath, Ms. Bhawna Sharma, Ms. Pooja AR, Mr. Ro- han C Jadhav, Ms.Asha A,”Autonomous Garbage Collecting Robot Wall- E”,IJSRD - International Journal for Scientific Research Development, ISSN (online): 2321-0613. [9] G. Sivasankar,B. Durgalakshmi,K .Seyatha,”Autonomous Trash Collect- ing Robot” International Journal of Engineering Research Technology (IJERT) ISSN: 2278-0181. [10] Saurav Kumar , Drishti Yadav , Himanshu Gupta , Om Prakash Verma , Irshad Ahmad Ansari and Chang Wook Ahn , Department of Instrumentation and Control Engineering, Dr. B. R. Ambedkar National Institute of Technology Jalandhar, Punjab 144011, India , Novel YOLOv3 Algorithm-Based Deep Learning Approach for Waste Segregation: Towards Smart Waste Management, Electronics 2021, 10, 14 https://dx.doi.org/10.3390/ electronics100100141. Authorized licensed use limited to: Sharda University. Downloaded on May 24,2022 at 06:03:49 UTC from IEEE Xplore. Restrictions apply.